Fully Automated Placental Volume Quantification From 3D Ultrasound for Prediction of Small‐for‐Gestational‐Age Infants
Open accessAbstract
Objectives: Early placental volume (PV) has been associated with small‐for‐gestational‐age infants born under the 10th/5th centiles (SGA10/SGA5). Manual or semiautomated PV quantification from 3D ultrasound (3DUS) is time intensive, limiting its incorporation into clinical care. We devised a novel convolutional neural network (CNN) pipeline for fully automated placenta segmentation from 3DUS images, exploring the association between the calculated PV and SGA. Methods: Volumes of 3DUS obtained from singleton pregnancies at 11–14 weeks' gestation were automatically segmented by our CNN pipeline trained and tested on 99/25 images, combining two 2D and one 3D models with downsampling/upsampling architecture. The PVs derived from the automated segmentations (PV CNN) were used to train multivariable logistic‐regression classifiers for SGA10/SGA5. The test performance for predicting SGA was compared to PVs obtained via the semiautomated VOCAL (GE‐Healthcare) method (PV VOCAL). Results: We included 442 subjects with 37 (8.4%) and 18 (4.1%) SGA10/SGA5 infants, respectively. Our segmentation pipeline achieved a mean Dice score of 0.88 on an independent test‐set. Adjusted models including PV CNN or PV VOCAL were similarly predictive of SGA10 (area under curve [AUC]: PV CNN = 0.780, PV VOCAL = 0.768). The addition of PV CNN to a clinical model without any PV included (AUC = 0.725) yielded statistically significant improvement in AUC (P < .05); whereas PV VOCAL did not (P = .105). Moreover, when predicting SGA5, including the PV CNN(0.897) brought statistically significant improvement over both the clinical model (0.839,P = .015) and the PV VOCAL model (0.870,P = .039). Conclusions: First trimester PV measurements derived from our CNN segmentation pipeline are significantly associated with future SGA. This fully automated tool enables the incorporation of including placental volumetric biometry into the bedside clinical evaluation as part of a multivariable prediction model for risk stratification and patient counseling.
AI snapshot
Written from the paper’s full text.
TL;DR
A fully automated CNN pipeline segmented the placenta from first-trimester 3D ultrasound and calculated placental volume (PV) with a mean Dice score of 0.88 on an independent test set. In a 442-subject cohort, CNN-derived PV improved prediction of SGA10 and SGA5 over a clinical model, and for SGA5 it outperformed semi-automated VOCAL-derived PV.
Key findings
- The CNN segmentation pipeline achieved a mean Dice score of 0.8829±0.0457 on the N=25 independent test set and 0.8721±0.0591 in 4-fold cross-validation; test-set mean boundary distance was 1.38±0.72 mm and mean Hausdorff distance was 18.60±11.46 mm.
- In the N=442 study set, there were 37 SGA10 (8.4%) and 18 SGA5 (4.1%) infants. For SGA10, adjusted models with PVCNN or PVVOCAL had similar AUCs (0.780 vs 0.768); adding PVCNN to the clinical model (AUC=0.725) significantly improved AUC (P=0.047), whereas PVVOCAL did not (P=0.105).
- For SGA5, the clinical model had AUC=0.839; adding PVCNN improved AUC to 0.897 (P=0.015), and PVCNN significantly outperformed PVVOCAL (0.870, P=0.039).
- PVCNN correlated strongly with manually labeled ground-truth PV (Pearson=0.908, R²=0.824), and paired automated PV measurements had ICC=0.867. Mean PVCNN was not significantly different from ground truth (78.7±29.7 cc vs 79.4±30.9 cc, P=0.878), while PVVOCAL was significantly lower (65.2±26.9 cc; P=0.003 vs ground truth, P=0.005 vs CNN).
Methods
- Secondary analysis of a prospective cohort of singleton pregnancies at 11–14 weeks’ gestation using GE Voluson E8 3DUS. The study used an N=442 SGA study set, an N=99 manually labeled training set, and an N=25 independent test set; manual ground truth was created with ITK-SNAP.
- The segmentation pipeline combined two 2D CNNs and one 3D CNN with downsampling/upsampling architecture, strided convolutions, no skip connections, and averaged probability maps thresholded to produce final segmentations. It was compared with semi-automated VOCAL and a vanilla 2D U-net baseline.
- SGA prediction used logistic regression with PVCNN or PVVOCAL plus clinical covariates (maternal age, nulliparity, race, BMI, tobacco use, chronic hypertension, pregestational diabetes), with backward elimination and ROC/AUC comparisons via DeLong’s test.
Limitations
- The paper states that the manually labeled dataset was limited (N=99), raising overfitting concerns; all images came from the same team, equipment, and single institution, so performance may drop on external data. It also notes that manual quality control removed 62 poor images, but the N=442 set still included below-average quality images causing oversegmentation, and that there is no clear mathematical ground truth for placenta segmentation due to manual variability.
- The paper notes the vanilla 2D U-net baseline was not extensively hyperparameter-tuned, so its Dice score is not directly comparable to the proposed pipeline.
- Evident from the text: SGA outcomes were rare (37 SGA10 and 18 SGA5), and there was no external validation cohort; VOCAL volumes were unavailable for the N=25 independent test set.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
0 of 37 cited sources checked · 0 supported · 0 flagged
- Findingown data
A combination of two 2D and one 3D CNN maximized the Dice score compared to their separate scores.
“For our segmentation task, we found that a combination of two 2D and one 3D CNNs maximized the Dice score compared to their separate scores.”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
For SGA5, the PVVOCAL model had an AUC of 0.870.
“Moreover, when predicting SGA5, including the PVCNN (0.897) brought statistically significant improvement over both the clinical model (0.839, P=0.015), as well as the PVVOCAL model (0.870, P=0.039).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
Adding further models beyond the two 2D and one 3D CNNs did not bring significant improvement.
“Adding further models did not bring significant improvement (Fig. 2).”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
Automated placental volume for primary images correlated with manually labeled ground truth placental volumes with Pearson=0.908 and R2=0.824.
“Figure 5c and 5d shows the correlation between the automated PV for the primary images and their manually labeled ground truth PVs (prediction for 1st scan vs GT for 1st scan, Pearson=0.908, R2=0.824).”
Quote in the paperSubstantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
Adding PVVOCAL to the clinical model did not improve AUC for SGA10 with statistical significance (P=0.105).
“The addition of PVCNN to a clinical model without any PV included (AUC=0.725) yielded statistically significant improvement in AUC (P<0.05); whereas, PVVOCAL did not (P=0.105).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
A clinical model without any placental volume had an AUC of 0.725 for SGA10.
“The addition of PVCNN to a clinical model without any PV included (AUC=0.725) yielded statistically significant improvement in AUC (P<0.05); whereas, PVVOCAL did not (P=0.105).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
For SGA5, including PVCNN improved over the PVVOCAL model with P=0.039.
“Moreover, when predicting SGA5, including the PVCNN (0.897) brought statistically significant improvement over both the clinical model (0.839, P=0.015), as well as the PVVOCAL model (0.870, P=0.039).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
In 4-fold cross-validation, the CNN pipeline's mean Hausdorff distance was 18.19±11.40 mm.
“The mean Hausdorff distance was 18.19±11.40 mm (min=4.24 mm, median=15.30 mm, max=47.98 mm).”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
Adding PVCNN to the clinical model improved AUC for SGA10 with statistical significance (P<0.05).
“The addition of PVCNN to a clinical model without any PV included (AUC=0.725) yielded statistically significant improvement in AUC (P<0.05); whereas, PVVOCAL did not (P=0.105).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
The CNN segmentation pipeline achieved a mean Dice score of 0.8829±0.0457 on the independent N=25 test set.
“Our segmentation pipeline achieved a mean Dice score of 0.88 on an independent test-set.”
- Also stated in: § Fully automated placenta segmentation pipeline · plain
“The mean Dice score on this test set was 0.8829±0.0457, which is in line with the cross-validation results reported above.”
Quote in the paperSubstantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Also stated in: § Fully automated placenta segmentation pipeline · plain
- Findingown data
The CNN pipeline's mean boundary distance on the N=25 test set was 1.38±0.72 mm.
“The mean boundary distance was 1.38±0.72 mm (min=0.51 mm, median=1.29 mm, max=3.36 mm).”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
For SGA5, the model including PVCNN had an AUC of 0.897.
“Moreover, when predicting SGA5, including the PVCNN (0.897) brought statistically significant improvement over both the clinical model (0.839, P=0.015), as well as the PVVOCAL model (0.870, P=0.039).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
In 4-fold cross-validation, the CNN pipeline's mean boundary distance was 1.54±1.31 mm.
“The mean boundary distance was 1.54±1.31 mm (min=0.37 mm, median=1.16 mm, max=8.33 mm), which roughly corresponds to 3 voxels.”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
Adjusted models including PVCNN or PVVOCAL were similarly predictive of SGA10, with AUCs of 0.780 for PVCNN and 0.768 for PVVOCAL.
“Adjusted models including PVCNN or PVVOCAL were similarly predictive of SGA10 (AUCs: PVCNN=0.780, PVVOCAL=0.768).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
Averaging the probability outputs of the two 2D models provided a Dice score improvement of 0.01.
“While they have similar performance separately, simply averaging their probability outputs provide a Dice score improvement of 0.01 (Fig.2).”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
In 4-fold cross-validation on the N=99 train set, the CNN pipeline achieved a mean Dice score of 0.8721±0.0591.
“Calculated over the test folds of the 4-fold CV setup, our fully automated CNN pipeline achieved a mean Dice result of 0.8721±0.0591 on the N=99 train set.”
Quote in the paperSubstantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
The CNN pipeline's mean Hausdorff distance on the N=25 test set was 18.60±11.46 mm.
“The mean Hausdorff distance was 18.60±11.46 mm (min=3.16 mm, median=14.50 mm, max=38.51 mm).”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
The vanilla 2D U-net baseline achieved a mean Dice score of 0.73 on the N=25 test set.
“When applied to the N=25 test set, this model achieved a mean Dice score of 0.73.”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Findingown data
First trimester placental volume measurements derived from the CNN segmentation pipeline are significantly associated with future SGA.
“First trimester placental volume measurements derived from our CNN segmentation pipeline are significantly associated with future SGA.”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
For SGA5, including PVCNN improved over the clinical model with P=0.015.
“Moreover, when predicting SGA5, including the PVCNN (0.897) brought statistically significant improvement over both the clinical model (0.839, P=0.015), as well as the PVVOCAL model (0.870, P=0.039).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
For SGA5, the clinical model had an AUC of 0.839.
“Moreover, when predicting SGA5, including the PVCNN (0.897) brought statistically significant improvement over both the clinical model (0.839, P=0.015), as well as the PVVOCAL model (0.870, P=0.039).”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Findingown data
At a 10% false positive ratio, the sensitivities for SGA10 prediction were 35%, 40%, and 35% for PVCNN, PVVOCAL, and no PV models, respectively.
“At a 10% false positive ratio, the sensitivities were 35%, 40%, and 35% respectively.”
Quote in the paperNot a substantial claim - Findingown data
In the whole cohort where PVVOCAL was available (N=516), PVVOCAL and PVCNN correlated (Pearson=0.676, R2=0.457).
“there was correlation between PVVOCAL and PVCNN (Pearson=0.676, R2=0.457, Fig. 6a).”
Quote in the paperNot a substantial claim - Findingown data
For SGA5, the clinical covariates most associated were tobacco use, chronic hypertension, pregestational diabetes, nulliparity, age, race, and BMI.
“The clinical covariates that were most associated with SGA5 were tobacco use, chronic hypertension, pregestational diabetes, nulliparity, age, race, and BMI.”
Quote in the paperNot a substantial claim - Findingown data
The Dice results from the proposed segmentation method were higher than those of other fully automatic methods.
“Our Dice results from the segmentation task were higher than other alternatives for fully automatic methods [11,12,13,14,17].”
Quote in the paperNot a substantial claim0 of 5 sources checked- cites [11] Looney P, Stevenson GN, Nicolaides KH, Plasencia W, Molloholli M, Natsis S, Collins SL: Automatic 3D ultrasound segmentation of the first trimester placenta using deep learning. IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017) 2017; 279– 282. source
Not checked in the free run.
- cites [12] Looney P, Stevenson GN, Nicolaides KH, Plasencia W, Molloholli M, Natsis S, Collins SL: Fully automated, real-time 3d ultrasound segmentation to estimate first trimester placental volume using deep learning. JCI Insight 2018; 3(11). PMC6124436 29875312 source
Not checked in the free run.
- cites [13] Oguz I, Pouch A, Yushkevich N, Wang H, Gee J, Schwartz N, Yushkevich P: Fully automated placenta segmentation from 3D ultrasound images. Perinatal, Preterm and Paediatric Image Analysis, PIPPI workshop, MICCAI 2016.
Not checked in the free run.
- cites [14] Oguz BU, Wang J, Yushkevich N, Pouch A, Gee J, Yushkevich PA, Schwartz N, Oguz I. Combining Deep Learning and Multi-atlas Label Fusion for Automated Placenta Segmentation from 3DUS. In Data Driven Treatment Response Assessment and Preterm, Perinatal, and Paediatric Image Analysis Springer, Cham. 2018; 138– 148. source
Not checked in the free run.
- cites [17] Yang X, Yu L, Li S, Wang X, Wang N, Qin J, Ni D, Heng PA. Towards automatic semantic segmentation in volumetric ultrasound. In International Conference on Medical Image Computing and Computer-Assisted Intervention Springer, Cham. 2017; 711– 719. source
Not checked in the free run.
- Findingown data
In the N=99 train set, mean PVVOCAL was significantly lower than PVCNN (p=0.005).
“mean PVVOCAL (65.2±26.9cc) was significantly lower than both PVGT (p=0.003) and PVCNN (p=0.005).”
Quote in the paperNot a substantial claim - Findingown data
In the N=99 train set, mean PVVOCAL (65.2±26.9cc) was significantly lower than ground truth (p=0.003).
“mean PVVOCAL (65.2±26.9cc) was significantly lower than both PVGT (p=0.003) and PVCNN (p=0.005).”
Quote in the paperNot a substantial claim - Findingown data
At a 20% false-positive cutoff, inclusion of PV measurements improved prediction of SGA5, with sensitivities of 83%, 83%, and 78% for PVCNN, PVVOCAL, and no PV models, respectively.
“Inclusion of PV measurements to the models yielded improvement in predicting SGA5 at a cutoff point of 20% false-positive, with sensitivities of 83%, 83%, and 78% for the PVCNN, PVVOCAL and no PV models, respectively.”
Quote in the paperSubstantial claim - Findingown data
In the whole cohort (N=516), mean PVVOCAL remained significantly lower than PVCNN (67.5±20.5 cc vs. 97.8±32.9 cc; p<0.0001).
“the mean PVVOCAL remained significantly lower than PVCNN (67.5±20.5 cc vs. 97.8±32.9 cc; p<0.0001, Fig. 6b).”
Quote in the paperNot a substantial claim - Findingown data
There was no significant difference in AUC between PVCNN and PVVOCAL models for SGA10 (P=0.605).
“There was no significant difference in the AUC between PVCNN and PVVOCAL (P=0.605).”
Quote in the paperSubstantial claim - Findingown data
Adding PVCNN significantly improved the AUC for SGA5 to 0.897 (P=0.015).
“there was significant improvement when adding PVCNN (AUC=0.897; P=0.015).”
Quote in the paperSubstantial claim - Findingown data
A prediction model using these clinical factors alone yielded an AUC of 0.839 for detecting SGA5.
“A prediction model using these factors alone yielded an AUC of 0.839 for detecting SGA5.”
Quote in the paperSubstantial claim - Findingown data
A prediction model using these clinical factors alone yielded an AUC of 0.725 for SGA10.
“A prediction model using these factors alone yielded an AUC of 0.725.”
Quote in the paperSubstantial claim - Findingown data
In the N=25 independent test set, automated segmentation volumes strongly correlated with ground truth (Pearson=0.908).
“There was strong correlation when comparing the ground truth volumes with volumes from our automated segmentation (PVGT =59.4±18.6cc, PVCNN =62.8±16.4cc, Pearson=0.908).”
Quote in the paperSubstantial claim - Findingown data
Adding PVCNN to the clinical model significantly increased the AUC for SGA10 to 0.780 (P=0.047).
“when adding PVCNN to the model, the AUC significantly increased to (0.780; P=0.047).”
Quote in the paperSubstantial claim - Findingown data
For SGA10, the clinical covariates most associated were tobacco use, chronic hypertension, pregestational diabetes, age, race, and BMI.
“The clinical covariates that were most associated with SGA10 were tobacco use, chronic hypertension, pregestational diabetes, age, race, and BMI.”
Quote in the paperNot a substantial claim - Findingown data
In the N=99 train set, mean placental volume from CNN (78.7±29.7cc) did not differ significantly from ground truth (79.4±30.9cc, P=0.878).
“there was no significant difference in mean PV values obtained by CNN (PVCNN 78.7±29.7cc) compared to the ground truth (PVGT =79.4±30.9cc, P=0.878).”
Quote in the paperSubstantial claim - Findingown data
There was significant improvement in AUC in the model using PVCNN compared to the PVVOCAL model for SGA5 (P=0.039).
“There was also significant improvement in AUC in the model using PVCNN compared to the PVVOCAL model (P=0.039).”
Quote in the paperSubstantial claim - Findingnone given
The AUC results from the logistic regression model were comparable to or higher than previous work [7,8].
“The AUC results from our logistic regression model were comparable to or higher than previous work [7,8].”
Quote in the paperNot a substantial claim0 of 2 sources checked- cites [7] Schwartz N, Coletta J, Srinivas S, Pessel C, Timor IE, Parry S, Salafia C: Placental volume measurements early in pregnancy predict adverse perinatal outcomes. American Journal of Obstetrics and Gynecology 2009; 201(6): 142– 143.19481723 source
Not checked in the free run.
- cites [8] Schwartz N, Quant HS, Sammel MD, Parry S: Macrosomia has its roots in early placental development. Placenta 2014; 35(9), 684– 690. 25064071 PMC4143475 source
Not checked in the free run.
- Findingown data
The mean boundary distance of the proposed method was worse than that of [12].
“Our Hausdorff distance was better than [11] and comparable to [12], while our mean boundary distance was worse than [12].”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [12] Looney P, Stevenson GN, Nicolaides KH, Plasencia W, Molloholli M, Natsis S, Collins SL: Fully automated, real-time 3d ultrasound segmentation to estimate first trimester placental volume using deep learning. JCI Insight 2018; 3(11). PMC6124436 29875312 source
Not checked in the free run.
- Findingown data
The CNNs usually err on the boundaries of the placenta.
“According to our observations, the CNNs correctly locate the central mass of the placenta on most cases, and usually err on the boundaries.”
Quote in the paperNot a substantial claim - Findingown data
The ability to detect >60% of SGA5 infants as early as 11-14 weeks without the need for blood tests or repeat visits is encouraging.
“While the ability to detect >60% of SGA5 infants as early as 11–14 weeks without the need for blood tests or repeat visits is encouraging, validating this on new, unseen data would further augment the confidence and utility of this approach.”
Quote in the paperSubstantial claim - Findingreasoning
Low quality images deteriorate the segmentation performance of the CNN models.
“Low quality images like these understandably deteriorate the segmentation performance of the CNN models.”
Quote in the paperNot a substantial claim - Findingown data
Prediction for a single image with the current fully CNN approach completes within 1-2 minutes.
“Our current fully CNN approach achieves the same segmentation performance, and prediction for a single image completes within 1–2 minutes.”
Quote in the paperNot a substantial claim - Findingown data
According to the authors' observations, the CNNs correctly locate the central mass of the placenta in most cases.
“According to our observations, the CNNs correctly locate the central mass of the placenta on most cases, and usually err on the boundaries.”
Quote in the paperNot a substantial claim - Findingown data
The Hausdorff distance of the proposed method was better than that of [11].
“Our Hausdorff distance was better than [11] and comparable to [12], while our mean boundary distance was worse than [12].”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [11] Looney P, Stevenson GN, Nicolaides KH, Plasencia W, Molloholli M, Natsis S, Collins SL: Automatic 3D ultrasound segmentation of the first trimester placenta using deep learning. IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017) 2017; 279– 282. source
Not checked in the free run.
- Findingown data
The Hausdorff distance of the proposed method was comparable to that of [12].
“Our Hausdorff distance was better than [11] and comparable to [12], while our mean boundary distance was worse than [12].”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [12] Looney P, Stevenson GN, Nicolaides KH, Plasencia W, Molloholli M, Natsis S, Collins SL: Fully automated, real-time 3d ultrasound segmentation to estimate first trimester placental volume using deep learning. JCI Insight 2018; 3(11). PMC6124436 29875312 source
Not checked in the free run.
- Findingown data
The current fully CNN approach achieves the same segmentation performance as the previous hybrid method.
“Our current fully CNN approach achieves the same segmentation performance, and prediction for a single image completes within 1–2 minutes.”
Quote in the paperSubstantial claim - Premisecitation
Abnormalities in gross placental development are linked to adverse pregnancy outcomes such as preeclampsia and intrauterine growth restriction.
“Mounting evidence [1,2,3] has linked abnormalities in gross placental development to such outcomes.”
Quote in the paperSubstantial claim0 of 3 sources checked- cites [1] Baptiste-Roberts K, Salafia CM, Nicholson WK, Duggan A, Wang NY, Brancati FL: Gross placental measures and childhood growth. The Journal of Maternal-Fetal & Neonatal Medicine 2009; 22(1): 13– 23. 19085212 PMC3713228 source
Not checked in the free run.
- cites [2] Barker DJ, Bull AR, Osmond C, Simmonds SJ: Fetal and placental size and risk of hypertension in adult life. British Medical Journal 1990; 301(6746): 259– 62. 2390618 PMC1663477 source
Not checked in the free run.
- cites [3] Biswas S, Ghosh SK: Gross morphological changes of placentas associated with intrauterine growth restriction of fetuses: A case control study. Early Human Development 2008; 84(6), 357– 362. 18093757 source
Not checked in the free run.
- Premisecitation
First trimester placental volume is associated with adverse pregnancy outcomes, including small-for-gestational-age infants born under the 10th or 5th centiles.
“Early placental volume (PV) has been associated with small-for-gestational-age infants born under the 10th/5th centiles (SGA10/SGA5).”
- Also stated in: § Introduction · boosted
“Advances in 3D ultrasound (3DUS) have allowed investigators to demonstrate associations between first trimester placental volume and clinical outcomes [5,6,7,8,9]”
Quote in the paperSubstantial claim0 of 5 sources checked- cites [5] Hafner E, Metzenbauer M, Stümpflen I, Waldhör T, Philipp K. First trimester placental and myometrial blood perfusion measured by 3D power Doppler in normal and unfavourable outcome pregnancies. Placenta 2010; 31(9): 756– 63. 20633928 source
Not checked in the free run.
- cites [6] Metzenbauer M, Hafner E, Schuchter K, Philipp K. First-trimester placental volume as a marker for chromosomal anomalies: preliminary results from an unselected population. Ultrasound in Obstetrics and Gynecology: The Official Journal of the International Society of Ultrasound in Obstetrics and Gynecology 2002; 19(3): 240– 242. 11896942 source
Not checked in the free run.
- cites [7] Schwartz N, Coletta J, Srinivas S, Pessel C, Timor IE, Parry S, Salafia C: Placental volume measurements early in pregnancy predict adverse perinatal outcomes. American Journal of Obstetrics and Gynecology 2009; 201(6): 142– 143.19481723 source
Not checked in the free run.
- cites [8] Schwartz N, Quant HS, Sammel MD, Parry S: Macrosomia has its roots in early placental development. Placenta 2014; 35(9), 684– 690. 25064071 PMC4143475 source
Not checked in the free run.
- cites [9] Schwartz N, Sammel MD, Leite R, Parry S. First-trimester placental ultrasound and maternal serum markers as predictors of small-for-gestational-age infants. American Journal of Obstetrics and Gynecology 2014; 211(3): 253– e1. 24607753 PMC4149853 source
Not checked in the free run.
- Also stated in: § Introduction · boosted
- Premisecitation
Early in utero placental size measured with 2D ultrasound has weak associations with adverse outcomes.
“Early, in utero placental size as measured with 2D ultrasound has demonstrated weak associations with adverse outcomes [4].”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [4] Schwartz N, Wang E, Parry S: Two-dimensional sonographic placental measurements in the prediction of small-for-gestational-age infants. Ultrasound in Obstetrics & Gynecology 2012; 40(6), 674– 679. 22331557 source
Not checked in the free run.
- Study design and conductown data
The N=442 SGA study set consists of subjects unseen by the neural networks during training.
“we used the N=442 unlabeled dataset, which consists of subjects unseen by the neural networks during training.”
Quote in the paperSubstantial claim - Interpretationreasoning
The fully automated tool enables incorporation of placental volumetric biometry into bedside clinical evaluation as part of a multi-variable prediction model for risk stratification and patient counseling.
“This fully automated tool enables the incorporation of including placental volumetric biometry into the bedside clinical evaluation as part of a multi-variable prediction model for risk stratification and patient counseling.”
Quote in the paperSubstantial claim- Rests on an assumption: The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
- Interpretationnone given
The placental volume measurements obtained at 11-14 weeks using the automated tool are reproducible.
“The placental volume measurements obtained at 11–14 weeks using this automated tool are validated compared to manual segmentation, reproducible, and significantly associated with the delivery of a small-for-gestational-age infant.”
Quote in the paperSubstantial claim - Interpretationown data
The placental volume measurements obtained at 11-14 weeks using the automated tool are validated compared to manual segmentation.
“The placental volume measurements obtained at 11–14 weeks using this automated tool are validated compared to manual segmentation, reproducible, and significantly associated with the delivery of a small-for-gestational-age infant.”
Quote in the paperSubstantial claim - Premisecitation
k-fold cross-validation is more robust to variance from random splitting of a small dataset than a single train/validate split.
“The k-fold CV setup has the advantage of being more robust to variance introduced during random splitting of a small dataset compared to a single train/validate split, since we are running k simultaneous models with identical architecture and hyper-parameters on k different train/validation splits [22,23].”
Quote in the paperSubstantial claim0 of 2 sources checked- cites [22] Blum A, Kalai A, Langford J. Beating the hold-out: Bounds for k-fold and progressive cross-validation. In Proceedings of the twelfth annual conference on Computational learning theory 1999; 203– 208. source
Not checked in the free run.
- cites [23] Kohavi R A study of cross-validation and bootstrap for accuracy estimation and model selection. In Ijcai 1995; 14(2): 1137– 1145.
Not checked in the free run.
- Interpretationown data
The placental volume measurements obtained at 11-14 weeks using the automated tool are significantly associated with the delivery of a small-for-gestational-age infant.
“The placental volume measurements obtained at 11–14 weeks using this automated tool are validated compared to manual segmentation, reproducible, and significantly associated with the delivery of a small-for-gestational-age infant.”
Quote in the paperSubstantial claim- Rests on an assumption: Placental size at birth correlates with neonatal birth weight.
- Interpretationreasoning
A perfect Dice score that predicts manual segmentation exactly in every subject is practically unattainable.
“Thus, a perfect Dice score that predicts the manual segmentation exactly in every subject is practically unattainable.”
Quote in the paperNot a substantial claim - Interpretationreasoning
The main advantage of the model is that apart from PV, it only uses easily obtainable data as features (tobacco use, chronic hypertension, pregestational diabetes, nulliparity, age, race, BMI).
“The main advantage of our model is that apart from PV, it only uses easily obtainable data as features (tobacco use, chronic hypertension, pregestational diabetes, nulliparity, age, race, BMI).”
Quote in the paperSubstantial claim - Interpretationown data
The random online data augmentations used during training were capable of preventing significant overfitting.
“These two additional checks strongly support our belief that the random online data augmentations used during training were capable of preventing significant overfitting.”
Quote in the paperSubstantial claim
- Classified as probably not a substantial claim. 22
- May combine two claims. 19
- The quote only partly states this claim. 2
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.
Findingown dataAlternative configurations produced cross-validation Dice scores between 0.80 and 0.83, poorer than the main pipeline's 0.87.
“using a Dice loss function (0.82), using a combined loss function of Dice and cross entropy losses weighted equally (0.83), implementing skip connections between the downsampling/upsampling paths (0.82), using max-pooling instead of strided convolutions (0.80), omitting drop-out layers (0.82), and omitting batch-normalization (0.81).”
- Also stated in: § Fully automated placenta segmentation pipeline · plain
“These alternatives were not used in the final configuration presented in this paper, since they produced poorer CV performance compared to our main pipeline (0.87).”
Quote in the paperNot a substantial claim- Rests on an assumption: Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
- Also stated in: § Fully automated placenta segmentation pipeline · plain
May combine two claims.
Findingown dataThe study included 442 subjects, with 37 (8.4%) SGA10 and 18 (4.1%) SGA5 infants.
“We included 442 subjects with 37 (8.4%) and 18 (4.1%) SGA10/SGA5 infants, respectively.”
- Also stated in: § Results · plain
“SGA10 was found in 8.4% (N=37); whereas, SGA5 was found in 4.1% (N=18).”
Quote in the paperNot a substantial claim- Also stated in: § Results · plain
May combine two claims.
Findingown dataFor subjects with paired scans, automated placental volume measurements had Pearson correlation 0.870, R2=0.757, and ICC 0.867.
“The Pearson coefficient was 0.870, and R2=0.757. The intraclass correlation coefficient (ICC) was 0.867.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataAdding PVVOCAL to the clinical model increased the AUC for SGA10 to 0.768, but the increase was not statistically significant (P=0.105).
“The AUC also increased when using PVVOCAL in the clinical model, although the increase did not reach statistical significance (0.768; P=0.105).”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataAdding PVVOCAL increased the AUC for SGA5 to 0.870, but was not statistically significant (P=0.092).
“Adding PVVOCAL (AUC=0.870; P=0.092) increased the AUC, but was not statistically significant.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataAt a 20% false positive ratio, the prediction of SGA10 was not significantly improved by including PV measurements, with sensitivities of 51%, 51%, and 54% for PVCNN, PVVOCAL, and no PV models, respectively.
“If we assume a 20% false positive ratio as a clinically appropriate cutoff point, the prediction of SGA10 was not significantly improved by including PV measurements with sensitivities for the PVCNN, PVVOCAL, and no PV models are 51%, 51%, and 54%, respectively.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataTotal time for segmentation of a new image, including pre-processing, CNN prediction, post-processing, and outputting segmentation, takes approximately 1 minute.
“Total time for segmentation of a new image (pre-processing, CNN prediction, post-processing, outputting segmentation) takes approximately 1 minute.”
Quote in the paperNot a substantial claimThe quote only partly states this claim.
Findingown dataAt a 10% false positive rate, the sensitivities for SGA5 prediction were 67%, 39%, and 44% for PVCNN, PVVOCAL, and no PV models, respectively.
“At a 10% false positive rate, the sensitivities were 67%, 39%, and 44% respectively.”
Quote in the paperSubstantial claimMay combine two claims.
FindingreasoningThe N=442 set included many images of below-average quality, which results in oversegmentation by the CNNs.
“The N=442 set still included many images of below-average quality, which results in oversegmentation by the CNNs.”
Quote in the paperNot a substantial claimMay combine two claims.
Premisenone givenCurrently available segmentation techniques for first-trimester placental volume from 3D ultrasound are laborious and require significant offline processing with manual input.
“but the currently available segmentation techniques are laborious and require significant offline processing with manual input.”
- Also stated in: § Objectives · plain
“Manual or semi-automated PV quantification from 3DUS is time-intensive, limiting its incorporation into clinical care.”
Quote in the paperSubstantial claim- Rests on an assumption: VOCAL requires offline manipulation and manual tracing of successive image slices within the 3D ultrasound volume set to segment the placenta.
- Also stated in: § Objectives · plain
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThere have been semi-automated and fully automated methods proposed for placental segmentation from 3D ultrasound, with varying results.
“There have been some semi-automated and fully automated methods proposed for the segmentation task, with varying results [10,11,12,13,14,15,16,17].”
Quote in the paperNot a substantial claim0 of 8 sources checked- cites [10] Collins SL, Stevenson GN, Noble JA, Impey L: Rapid Calculation of Standardized Placental Volume at 11 to 13 Weeks and the Prediction of Small for Gestational Age Babies. Ultrasound in Medicine & Biology 2013; 39(2), 253– 260. 23219036 source
Not checked in the free run.
- cites [11] Looney P, Stevenson GN, Nicolaides KH, Plasencia W, Molloholli M, Natsis S, Collins SL: Automatic 3D ultrasound segmentation of the first trimester placenta using deep learning. IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017) 2017; 279– 282. source
Not checked in the free run.
- cites [12] Looney P, Stevenson GN, Nicolaides KH, Plasencia W, Molloholli M, Natsis S, Collins SL: Fully automated, real-time 3d ultrasound segmentation to estimate first trimester placental volume using deep learning. JCI Insight 2018; 3(11). PMC6124436 29875312 source
Not checked in the free run.
- cites [13] Oguz I, Pouch A, Yushkevich N, Wang H, Gee J, Schwartz N, Yushkevich P: Fully automated placenta segmentation from 3D ultrasound images. Perinatal, Preterm and Paediatric Image Analysis, PIPPI workshop, MICCAI 2016.
Not checked in the free run.
- cites [14] Oguz BU, Wang J, Yushkevich N, Pouch A, Gee J, Yushkevich PA, Schwartz N, Oguz I. Combining Deep Learning and Multi-atlas Label Fusion for Automated Placenta Segmentation from 3DUS. In Data Driven Treatment Response Assessment and Preterm, Perinatal, and Paediatric Image Analysis Springer, Cham. 2018; 138– 148. source
Not checked in the free run.
- cites [15] Oguz I, Yushkevich N, Pouch A, Oguz BU, Wang J, Parameshwaran S, Gee J, Yushkevich PA, Schwartz N. Minimally interactive placenta segmentation from three-dimensional ultrasound images. Journal of Medical Imaging 2020; 7(1): 014004. 32118089 PMC7035878 source
Not checked in the free run.
- cites [16] Stevenson GN, Collins SL, Ding J, Impey L, Noble JA: 3-D Ultrasound Segmentation of the Placenta Using the Random Walker Algorithm: Reliability and Agreement. Ultrasound in Medicine & Biology 2015; 41(12): 3182– 3193. 26341043 source
Not checked in the free run.
- cites [17] Yang X, Yu L, Li S, Wang X, Wang N, Qin J, Ni D, Heng PA. Towards automatic semantic segmentation in volumetric ultrasound. In International Conference on Medical Image Computing and Computer-Assisted Intervention Springer, Cham. 2017; 711– 719. source
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe performance of a model in k-fold cross-validation is estimated as the mean of the k validation results.
“The performance of the model on such a setup is estimated to be the mean of the k validation results [19,20,21].”
Quote in the paperNot a substantial claim0 of 3 sources checked- cites [19] Allen DM. The relationship between variable selection and data agumentation and a method for prediction. Technometrics 1974; 16(1): 125– 7. source
Not checked in the free run.
- cites [20] Stone M Cross validatory choice and assessment of statistical predictions. Journal of the Royal Statistical Society: Series B (Methodological) 1974; 36(2): 111– 33. source
Not checked in the free run.
- cites [21] Stone M An asymptotic equivalence of choice of model by cross validation and Akaike’s criterion. Journal of the Royal Statistical Society: Series B (Methodological) 1977; 39(1): 44– 7. source
Not checked in the free run.
Classified as probably not a substantial claim.
Study design and conductnone givenModels were adjusted for potential confounders including maternal age, nulliparity, race, BMI, tobacco use, chronic hypertension, and pregestational diabetes.
“Models were also adjusted for potential confounders, including maternal age, nulliparity, race, BMI, tobacco use, chronic hypertension, and pregestational diabetes.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductown dataThe baseline U-net's performance is not directly comparable to the pipeline's 0.88 Dice score because it did not undergo extensive hyper-parameter tuning.
“However, it must be noted that we did not perform extensive hyper-parameter tuning with this model, so its performance is not directly comparable to the 0.88 Dice score of our pipeline.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductown dataThe CNNs were trained on an N=99 training set under a 4-fold cross-validation scheme for tuning the models.
“The CNNs are trained on an N=99 training set, under a 4-fold cross-validation scheme for tuning the models.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim. May combine two claims.
Study design and conductnone givenThe N=442 SGA study set, N=99 train set, and N=25 test set are all independent from each other.
“The N=442 SGA study set, the N=99 train set, and the N=25 test set are all independent from each other.”
Quote in the paperSubstantial claimMay combine two claims.
PremisecitationVolume data sets from the cohort were previously analyzed using the VOCAL semi-automated segmentation method and found to be significantly associated with SGA.
“Volume data sets were previously analyzed using VOCAL semi-automated segmentation method and found to be significantly associated with SGA.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [9] Schwartz N, Sammel MD, Leite R, Parry S. First-trimester placental ultrasound and maternal serum markers as predictors of small-for-gestational-age infants. American Journal of Obstetrics and Gynecology 2014; 211(3): 253– e1. 24607753 PMC4149853 source
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationEnsembling multiple CNN models is a common deep learning approach for improving prediction results.
“Ensembling multiple CNN models is a common deep learning approach for improving prediction results [27,28,29].”
Quote in the paperNot a substantial claim0 of 3 sources checked- cites [27] Dietterich TG. Ensemble methods in machine learning. International workshop on multiple classifier systems Springer, Berlin, Heidelberg 2000; 1– 15. source
Not checked in the free run.
- cites [28] Drucker H, Cortes C, Jackel LD, LeCun Y, Vapnik V. Boosting and other ensemble methods. Neural Computation 1994; 6(6): 1289– 301. source
Not checked in the free run.
- cites [29] Opitz D, Maclin R. Popular ensemble methods: An empirical study. Journal of artificial intelligence research 1999; 11: 169– 98. source
Not checked in the free run.
Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.
InterpretationreasoningEarly identification of pregnancies at risk for fetal growth disturbance can inform clinicians and allow for increased surveillance in the third trimester.
“Early identification of pregnancies at risk for fetal growth disturbance can inform clinicians and allow for increased surveillance in the third trimester.”
Quote in the paperNot a substantial claim- Rests on an assumption: Placental size at birth correlates with neonatal birth weight.
May combine two claims.
Interpretationown dataThe tool has the potential to be used at the bedside because the whole pipeline completes within minutes and requires no input from the user.
“This tool has the potential to be used at the bedside, since the whole pipeline for a new image completes within minutes, and more importantly, requires no input from the user.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationnone givenThe proposed method provides a complete and fully automated pipeline that takes in 3DUS image and clinical data, and outputs placenta segmentation, placenta volume, and probabilities for SGA5 and SGA10 outcomes.
“Our proposed method provides a complete and fully automated pipeline that takes in 3DUS image and clinical data, and outputs placenta segmentation, placenta volume, and probabilities for SGA5 and SGA10 outcomes.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductown dataThe human study was approved by the University of Pennsylvania Institutional Review Board and all adult participants provided written informed consent.
“This human study was approved by the University of Pennsylvania Institutional Review Board - approval: 811129. All adult participants provided written informed consent to participate in this study.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S2 (checkable).
InterpretationreasoningThe model can be introduced as an assisting tool to most clinical settings.
“This ensures that it can be introduced as an assisting tool to most clinical settings.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S2 (checkable).
InterpretationreasoningThere is no clear-cut, mathematical ground truth for placenta segmentation due to variability and human subjectivity in manual segmentation.
“there is no clear-cut, mathematical ground truth for placenta segmentation as there is variability and human subjectivity factored into the tedious process of manually segmenting 3DUS images with varying resolution and often unclear borders between placenta and myometrium.”
Quote in the paperNot a substantial claimMay combine two claims.
Interpretationown dataThe approach generalizes well to the larger, unlabeled SGA study set (N=442), as evidenced by performing better than VOCAL for SGA10 and SGA5 prediction.
“our approach generalizes well to the larger, unlabeled SGA study set (N=442), as evidenced by the fact that it performs better than VOCAL for SGA10 and SGA5 prediction.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningIt is possible that performance may drop when applied to data obtained by other teams/equipment if there are significant differences in raw 3DUS images due to methodology or hardware factors.
“it could be possible to witness some performance drop when applied to data obtained by other teams/equipment, if there are significant differences in the raw 3DUS images due to methodology or hardware factors.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S2 (checkable). May combine two claims.
Interpretationnone givenAutomated placental segmentation and volume measurement would enable the inclusion of placental biometry in clinical settings and more robust prospective research to improve risk stratification early in pregnancy.
“Automated placental segmentation and volume measurement would enable the inclusion of placental biometry in both a clinical setting and to conduct more robust prospective research to improve risk stratification early in pregnancy.”
Quote in the paperNot a substantial claim- Rests on an assumption: Quantitative assessment of the placenta is not part of routine clinical care, largely due to the lack of tools that allow this task to be accomplished in a timely and simplified manner.
Classified as probably not a substantial claim. May combine two claims.
Study design and conductreasoningAn ensemble of 3 models was employed to overcome the limitations of the relatively small dataset and to achieve an increase in performance and robustness compared to a single model.
“We employed an ensemble of 3 models in order to overcome the limitations of our relatively small dataset, and to achieve an increase in performance and robustness compared to a single model.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductcitationA manual quality control round was performed to eliminate 62 extremely poor images from the initial study cohort while building the train, test and SGA datasets.
“we performed a manual quality control round to eliminate 62 extremely poor images from our initial study cohort [9], while building the train, test and SGA datasets.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [9] Schwartz N, Sammel MD, Leite R, Parry S. First-trimester placental ultrasound and maternal serum markers as predictors of small-for-gestational-age infants. American Journal of Obstetrics and Gynecology 2014; 211(3): 253– e1. 24607753 PMC4149853 source
Not checked in the free run.
Probably not a substantial claim: may not pass S2 (checkable).
InterpretationreasoningThe goal of incorporating direct assessment of placental development into the clinical management of pregnancy is achievable by creating simplified and automated tools.
“Overall, by creating simplified and automated tools, the goal of incorporating direct assessment of placental development into the clinical management of pregnancy is achievable.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductnone givenAll images in the datasets were obtained by the same team and equipment in a single institution.
“all images in our datasets were obtained by the same team and equipment in a single institution.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).
Interpretationnone givenContinued refinement of the tool can enable a more robust assessment of other morphometric features such as placental surface area and shape.
“continued refinement of this tool can enable a more robust assessment of other morphometric features such as placental surface area, shape.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).
InterpretationreasoningAn automated placental segmentation tool can be extended to aid in volumetric quantification or mapping of Doppler signal as novel ultrasound technologies are developed.
“as novel ultrasound-based technologies, such as microvascular Doppler, are developed, an automated placental segmentation tool can be extended to aid in volumetric quantification or mapping of the Doppler signal as well.”
Quote in the paperNot a substantial claimMay combine two claims.
InterpretationreasoningThe model does not require additional blood tests, imaging or repeat visits.
“Therefore, it does not require additional blood tests, imaging or repeat visits.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Study design and conductown dataThe current approach was achieved with a very limited dataset of 99 images.
“We would also like to note this was achieved with a very limited dataset of 99 images.”
Quote in the paperSubstantial claim
The paper relies on this without support.
Automated placental quantification from 3D ultrasound is demanding because of speckle noise, weak early-pregnancy contrast, and variable placental size, shape, and position.
“Automated quantification of the placenta from 3DUS data is a demanding task. 3DUS images are prone to high levels of speckle noise. Especially in early pregnancy, the contrast between the placenta and uterine tissue is often rather weak. The size and shape of the placenta, as well as its position with respect to the amniotic sac, are all extremely variable.”
The paper relies on this without support.
VOCAL requires offline manipulation and manual tracing of successive image slices within the 3D ultrasound volume set to segment the placenta.
“However, VOCAL requires offline manipulation and manual tracing of successive image slices within the 3DUS volume set in order to segment out the organ of interest.”
The paper relies on this without support.
Manual segmentation using ITK-SNAP provides accurate ground truth for placental volume.
“99 3DUS volume sets from 99 subjects were manually segmented using the ITK-SNAP software (www.itksnap.org). This annotated dataset is used to train the 3 CNNs.”
The paper relies on this without support.
The CNN pipeline's segmentation performance on the N=25 independent test set generalizes to the N=442 SGA study set.
“After training the three final CNNs on the N=99 train set, these models were used to produce placenta segmentations for the N=442 study set which did not have manual segmentations, and placental volume (PVCNN) are automatically calculated for the study set from these segmentations.”
The paper relies on this without support.
Placental size at birth correlates with neonatal birth weight.
“Placental size at birth is known to correlate with neonatal birth weight.”
The paper relies on this without support.
Quantitative assessment of the placenta is not part of routine clinical care, largely due to the lack of tools that allow this task to be accomplished in a timely and simplified manner.
“However, quantitative assessment of the placenta is not part of routine clinical care, largely due to the lack of tools that allow this task to be accomplished in a timely and simplified manner.”
The paper relies on this without support.
Semi-automated tools, such as VOCAL, require manual segmentation of multiple image slices to obtain an estimated volume.
“However, this technique requires manual segmentation of multiple image slices to obtain an estimated volume.”
The paper relies on this without support.
Automated segmentation of the placenta is a demanding task.
“Automated segmentation of the placenta is a demanding task.”
The paper relies on this without support.
Interactive methods are subjective and inclined to introduce intra- and inter-observer variability.
“These interactive methods are subjective and inclined to introduce intra- and inter-observer variability.”
The paper relies on this without support.
Deep learning methods excel with large amounts of data.
“Since deep learning methods are known to excel with large amounts of data, obtaining more training images could likely further improve the Dice score of the CNN models.”
The paper relies on this without support.
Manual annotation is a time-intensive process that requires the labor of an expert.
“Manual annotation is a time-intensive process that requires the labor of an expert.”
The paper relies on this without support.
Training the models is a one time process.
“In any case, training the models is a one time process.”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Schwartz et al. and Looney et al. presented an effective model to automatically assess placenta volumes from 2D and 3D images in first trimester.
“Schwartz et al. and Looney et al. presented an effective model to automatically assess placenta volumes from 2D and 3D images in first trimester [112,113].”
From Evolving the Era of 5D Ultrasound? A Systematic Literature Review on the Applications for Artificial Intelligence Ultrasound Imaging in Obstetrics and Gynecology (2023) · cites this paper as [112]
The Introduction and Discussion describe a fully automated method that calculates first-trimester placental volume from 3D ultrasound images, with measurements validated against manual segmentation. The Methods say the pipeline combines 2D and 3D models, but its input is 3D ultrasound volumes; this does not establish that it assesses volume from separately acquired 2D images. This verdict addresses the attribution to Schwartz et al., not the separate citation to Looney et al. (LLM, conf=0.88)
Experimental — accuracy not yet measured
The citation lists haven’t been fetched yet
References 35 · Cited by 11
Research works printed but not linked: 1
- OguzI PouchA YushkevichN et al.Fully automated placenta segmentation from 3D ultrasound images. Paper presented at: Perinatal Preterm and Paediatric Image Analysis PIPPI Workshop MICCAI 2016.
- Authors
- Schwartz, N., Oguz, I., Wang, J., Pouch, A., Yushkevich, N., Parameshwaran, S., Gee, J., Yushkevich, P., & Oguz, B.
- Affiliations
- University of Pennsylvania (US)
- Department of EECS Vanderbilt University Nashville TN USA
- Corresponding author
- Oguz, Baris
- Venue
- Journal of Ultrasound in Medicine
- Publisher
- Wiley
- Published
- 2021-09-23
- Volume
- 41
- Issue
- 6
- Pages
- 1509-1524
- ISSN
- 0278-4297 (ISSN-L)
- 1550-9613
- PMC
- PMC8940735
- OpenAlex
- W3201127886
- Cited by
- 11
- References
- 35
- Citations by year
- 2026: 3 · 2025: 1 · 2024: 3 · 2023: 2 · 2022: 2
- FWCI
- 0.63
- Type
- journal article
- Language
- en
- Open access
- green
- www.ncbi.nlm.nih.gov
- Locations
- Journal of Ultrasound in Medicine (publishedVersion)
- PubMed (publishedVersion)
- PubMed Central (acceptedVersion)
- Funders
- Eunice Kennedy Shriver National Institute of Child Health and Human Development (U01 HD087180)
- National Institutes of Health (F32 HL119010, R01 EB017255, R01 NS094456)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- placental volume, small for gestational age, 3D ultrasound, convolutional neural networks, first trimester, risk stratification, VOCAL, singleton pregnancy, automated image analysis, logistic regression prediction model, patient counseling
- MeSH terms
- Female, Gestational Age, Humans, Infant, Newborn, Infant, Small for Gestational Age, Placenta, Pregnancy, Pregnancy Trimester, First, Ultrasonography, Prenatal
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
