CoChat

Development of an AI model for DILI-level prediction using liver organoid brightfield images

Open access
Tan, Shiyi, Ding, Yan, Wang, Wei, Rao, Jianhua, Cheng, Feng, Zhang, Qiuyin, Xu, Tingting, Hu, Tianmu, Hu, Qinyi, Ye, Ziliang, Yan, Xiaopeng, Wang, Xiaowei, Li, Mingyue, Xie, Peng, Chen, Zaozao, Liang, Geyu, Pu, Yuepu, Zhang, Juan, Gu, Zhongze · 2025 · Communications Biology

The DOI matches its Crossref record; 35 of the 35 references it cites were found in a registry or an index.

Abstract

AI image processing techniques hold promise for clinical applications by enabling analysis of complex status information from cells. Importantly, real-time brightfield imaging has advantages of informativeness, non-destructive nature, and low cost over fluorescence imaging. Currently, human liver organoids (HLOs) offer an alternative to animal models due to their excellent physiological recapitulation including basic functions and drug metabolism. Here we show a drug-induced liver injury (DILI) level prediction model using HLO brightfield images (DILITracer) considering that DILI is the major causes of drug withdrawals. Specifically, we utilize BEiT-V2 model, pretrained on 700,000 cell images, to enhance 3D feature extraction. A total of 30 compounds from FDA DILIrank are selected (classified into Most-, Less-, and No-DILI) to activate HLOs and corresponding brightfield images are collected at different time series and z-axis. Our computer vision model based on image-spatial-temporal coding layer excavates fully spatiotemporal information of continuously captured images, links HLO morphology with DILI severity, and final output DILI level of compounds. DILITracer achieves an overall accuracy of 82.34%. To our knowledge, this is the first model to output ternary classification of hepatotoxicity. Overall, DILITracer, using clinical data as an endpoint categorization label, offers a rapid and effective approach for screening hepatotoxic compounds.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s42003-025-08205-6

Cites

35 of 35 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 The workflow of the development of the DILI-level prediction model. a The construction of the DILI experimental platform based on human liver organoids for collecting daily brightfield images treated with different compounds under different z-axis. b The establishment of DILI-level prediction model using different brightfield images of liver organoids with spatiotemporal information. c The application of AI model for predicting DILI level of compounds with unknown hepatoxicity.
Fig. 1 The workflow of the development of the DILI-level prediction model. a The construction of the DILI experimental platform based on human liver organoids for collecting daily brightfield images treated with different compounds under different z-axis. b The establishment of DILI-level prediction model using different brightfield images of liver organoids with spatiotemporal information. c The application of AI model for predicting DILI level of compounds with unknown hepatoxicity.CC BY · PubMed Central
Fig. 2 DILI toxicity testing platform based on human liver organoid models. a Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in human liver organoids exposed to Chlorpheniramine (with a label of No-DILI levels) from Day 0 to Day 3. b ALB inhibition (P = 0.6262, t = 0.5268) and ATP inhibition (P = 0.8040, t = 0.2652) of liver injury in human liver organoids exposed to Chlorpheniramine at end of Day 3. c Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in human liver organoids liver organoids exposed to Simvastatin (with a label of Less-DILI levels) from Day 0 to Day 3. d ALB inhibition (P = 0.0844, t = 2.285) and ATP inhibition (P = 0.0036, t = 6.110) of liver injury in human liver organoids exposed to Simvastatin at end of Day 3. e Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in human liver organoids exposed to Gefitinib (with a label of Most-DILI levels) from Day 0 to Day 3. f ALB inhibition (P = 0.0191, t = 3.800) and ATP inhibition (P < 0.0001, t = 88.86) of liver injury in human liver organoids exposed to Gefitinib at end of Day 3. Scale bar = 100 μm. *P < 0.05; **P < 0.01; ****P < 0.0001. Error bars represent mean ± SD across n = 3 biologically independent organoid samples. Effect sizes (Cohen’s d) and 95% confidence intervals (CIs) are provided in Supplementary Table 5.
Fig. 2 DILI toxicity testing platform based on human liver organoid models. a Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in human liver organoids exposed to Chlorpheniramine (with a label of No-DILI levels) from Day 0 to Day 3. b ALB inhibition (P = 0.6262, t = 0.5268) and ATP inhibition (P = 0.8040, t = 0.2652) of liver injury in human liver organoids exposed to Chlorpheniramine at end of Day 3. c Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in human liver organoids liver organoids exposed to Simvastatin (with a label of Less-DILI levels) from Day 0 to Day 3. d ALB inhibition (P = 0.0844, t = 2.285) and ATP inhibition (P = 0.0036, t = 6.110) of liver injury in human liver organoids exposed to Simvastatin at end of Day 3. e Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in human liver organoids exposed to Gefitinib (with a label of Most-DILI levels) from Day 0 to Day 3. f ALB inhibition (P = 0.0191, t = 3.800) and ATP inhibition (P < 0.0001, t = 88.86) of liver injury in human liver organoids exposed to Gefitinib at end of Day 3. Scale bar = 100 μm. *P < 0.05; **P < 0.01; ****P < 0.0001. Error bars represent mean ± SD across n = 3 biologically independent organoid samples. Effect sizes (Cohen’s d) and 95% confidence intervals (CIs) are provided in Supplementary Table 5.CC BY · PubMed Central
Fig. 3 DILI toxicity testing platform based on HepG2 spheroid models. a Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in HepG2 spheroids exposed to Hydrocortisone (with a label of No-DILI levels) from Day 0 to Day 3. b ALB inhibition (P = 0.0073, t = 5.039) and ATP inhibition (P = 0.0080, t = 4.900) of liver injury in HepG2 spheroids exposed to Hydrocortisone at end of Day 3. c Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in HepG2 spheroids exposed to Cyclophosphamide (with a label of Less-DILI levels) from Day 0 to Day 3. d ALB inhibition (P = 0.0045, t = 5.750) and ATP inhibition (P = 0.0080, t = 4.899) of liver injury in HepG2 spheroids exposed to Cyclophosphamide at end of Day 3. e Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in HepG2 spheroids exposed to Travafloxacin (with a label of Most-DILI levels) from Day 0 to Day 3. f ALB inhibition (P = 0.0564, t = 2.661) and ATP inhibition (P < 0.0001, t = 21.46) of liver injury in HepG2 spheroids exposed to Travafloxacin at end of Day 3. Scale bar = 100 μm. **P < 0.01; ****P < 0.0001. Error bars represent mean ± SD across n = 3 biologically independent organoid samples. Cohen’s d and 95% CI are provided in Supplementary Table 6.
Fig. 3 DILI toxicity testing platform based on HepG2 spheroid models. a Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in HepG2 spheroids exposed to Hydrocortisone (with a label of No-DILI levels) from Day 0 to Day 3. b ALB inhibition (P = 0.0073, t = 5.039) and ATP inhibition (P = 0.0080, t = 4.900) of liver injury in HepG2 spheroids exposed to Hydrocortisone at end of Day 3. c Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in HepG2 spheroids exposed to Cyclophosphamide (with a label of Less-DILI levels) from Day 0 to Day 3. d ALB inhibition (P = 0.0045, t = 5.750) and ATP inhibition (P = 0.0080, t = 4.899) of liver injury in HepG2 spheroids exposed to Cyclophosphamide at end of Day 3. e Brightfield morphology changes (12 original images and 1 stacked image) of liver injury in HepG2 spheroids exposed to Travafloxacin (with a label of Most-DILI levels) from Day 0 to Day 3. f ALB inhibition (P = 0.0564, t = 2.661) and ATP inhibition (P < 0.0001, t = 21.46) of liver injury in HepG2 spheroids exposed to Travafloxacin at end of Day 3. Scale bar = 100 μm. **P < 0.01; ****P < 0.0001. Error bars represent mean ± SD across n = 3 biologically independent organoid samples. Cohen’s d and 95% CI are provided in Supplementary Table 6.CC BY · PubMed Central
Fig. 4 Evaluation of the predictive performance of human liver organoids and HepG2 spheroids image-only models. a, b Confusion matrix of DILI classifier based on the human liver organoid and HepG2 spheroid dataset. c, d Indexes of overall accuracy, recall, specificity, precision, and F1 score of DILI classifier based on the human liver organoid and HepG2 spheroid dataset. e, f Receiver operating characteristic (ROC) curves of the human liver organoid and HepG2 spheroid image-only model. Green, yellow, and blue curves represent vNo-, vLess-, and vMost-DILI-Concern labels, respectively. AUC values are labeled for each curve.
Fig. 4 Evaluation of the predictive performance of human liver organoids and HepG2 spheroids image-only models. a, b Confusion matrix of DILI classifier based on the human liver organoid and HepG2 spheroid dataset. c, d Indexes of overall accuracy, recall, specificity, precision, and F1 score of DILI classifier based on the human liver organoid and HepG2 spheroid dataset. e, f Receiver operating characteristic (ROC) curves of the human liver organoid and HepG2 spheroid image-only model. Green, yellow, and blue curves represent vNo-, vLess-, and vMost-DILI-Concern labels, respectively. AUC values are labeled for each curve.CC BY · PubMed Central
Fig. 5 Superiority analysis and ability validation of human liver organoids image-only model. a Immunofluorescence detection of ZO-1, BSEP, HNF4a, DES, CD31, CD68 in human liver organoids. Scale bar = 50 μm. b Expression level of CYP3A4 (P = 0.0175, t = 3.900), CYP1A2 (P = 0.0022, t = 7.022), CYP2D6 (P = 0.0021, t = 7.108), CYP2E1 (P = 0.0010, t = 8.505), CYP2C9 (P < 0.0001, t = 20.51), and CYP2C19 (P = 0.0031, t = 6.374) between HepG2 spheroids and human liver organoids. *P < 0.05, ** P < 0.01; ****P < 0.0001. Error bars represent mean ± SD across n = 3 biologically independent organoid samples. Cohen’s d and 95% CI are provided in Supplementary Table 7. c Radar plot visualizing the confusion matrix in ablation experiments based on spatial dimensions and receiver operating characteristic (ROC) curves of human liver organoids image only model using Day0–Day1 images. d Radar plot visualizing the confusion matrix in ablation experiments based on temporal dimensions and ROC curves of human liver organoids image only model without the structure of the spatial coding layer. Green, yellow, and blue curves represent vNo-, vLess-, and vMost-DILI-Concern labels, respectively. e Attention heatmap of Gefitinib. z-axis (top image) = 135,650; z-axis (bottom image) = 135,950; f attention heatmap of Chlorphenamine. z-axis (top image) = 136,500; z-axis (bottom image) = 136,900; the red and the blue areas in the heatmap indicate the areas with higher and lower attention weights, respectively. The small boxes mark the areas that are in focus precisely under each visual field. The large red boxes mark the time period with higher activation of attention weights. Scale bar = 100 μm.
Fig. 5 Superiority analysis and ability validation of human liver organoids image-only model. a Immunofluorescence detection of ZO-1, BSEP, HNF4a, DES, CD31, CD68 in human liver organoids. Scale bar = 50 μm. b Expression level of CYP3A4 (P = 0.0175, t = 3.900), CYP1A2 (P = 0.0022, t = 7.022), CYP2D6 (P = 0.0021, t = 7.108), CYP2E1 (P = 0.0010, t = 8.505), CYP2C9 (P < 0.0001, t = 20.51), and CYP2C19 (P = 0.0031, t = 6.374) between HepG2 spheroids and human liver organoids. *P < 0.05, ** P < 0.01; ****P < 0.0001. Error bars represent mean ± SD across n = 3 biologically independent organoid samples. Cohen’s d and 95% CI are provided in Supplementary Table 7. c Radar plot visualizing the confusion matrix in ablation experiments based on spatial dimensions and receiver operating characteristic (ROC) curves of human liver organoids image only model using Day0–Day1 images. d Radar plot visualizing the confusion matrix in ablation experiments based on temporal dimensions and ROC curves of human liver organoids image only model without the structure of the spatial coding layer. Green, yellow, and blue curves represent vNo-, vLess-, and vMost-DILI-Concern labels, respectively. e Attention heatmap of Gefitinib. z-axis (top image) = 135,650; z-axis (bottom image) = 135,950; f attention heatmap of Chlorphenamine. z-axis (top image) = 136,500; z-axis (bottom image) = 136,900; the red and the blue areas in the heatmap indicate the areas with higher and lower attention weights, respectively. The small boxes mark the areas that are in focus precisely under each visual field. The large red boxes mark the time period with higher activation of attention weights. Scale bar = 100 μm.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 35 references

Research works printed but not linked: 1

  1. Dosovitskiy, A. et al. An image is worth 16×16 words: Transformers for image recognition at scale. Int. Conf. Learn. Represent. (2021).

Cited by

See all 26

Metadata

Authors
Tan, S., Ding, Y., Wang, W., Rao, J., Cheng, F., Zhang, Q., Xu, T., Hu, T., Hu, Q., Ye, Z., Yan, X., Wang, X., Li, M., Xie, P., Chen, Z., Liang, G., Pu, Y., Zhang, J., & Gu, Z.
Affiliations
  • Southeast University (CN)
  • Chinese Academy of Medical Sciences & Peking Union Medical College (CN)
  • State Key Laboratory of Digital Medical Engineering (CN)
Corresponding author
  • Zhang, Juan
  • Gu, Zhongze
Venue
Communications Biology
Published
2025-06-07
Volume
8
Issue
1
Pages
886-886
ISSN
2399-3642 (ISSN-L)
OpenAlex
W4411080647
Cited by
26
References
33
Citations by year
2026: 22 · 2025: 4
FWCI
25.08
Type
journal article
Language
en
Open access
Licence
Locations
  • Communications Biology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Government of Jiangsu Province
  • Natural Science Foundation of Jiangsu Province (BK20222008)
  • Nanjing Medical University
Topics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Pharmacology › Drug-Induced Hepatotoxicity and Protection
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Biophysics › Cell Image Analysis Techniques
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
Keywords
drug-induced liver injury, human liver organoids, bright-field microscopy, hepatotoxicity prediction, organoid morphology
MeSH terms
Artificial Intelligence, Humans, Image Processing, Computer-Assisted, Liver, Organoids, Chemical and Drug Induced Liver Injury

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.

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Development of an AI model for DILI-level prediction using liver organoid brightfield images. https://app.cochat.ai/papers/10.1038/s42003-025-08205-6
The paper
Tan, S., Ding, Y., Wang, W., Rao, J., Cheng, F., Zhang, Q., Xu, T., Hu, T., Hu, Q., Ye, Z., Yan, X., Wang, X., Li, M., Xie, P., Chen, Z., Liang, G., Pu, Y., Zhang, J., & Gu, Z. (2025). Development of an AI model for DILI-level prediction using liver organoid brightfield images. Communications Biology. https://doi.org/10.1038/s42003-025-08205-6

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.