CoChat
Sign Up
Cited by 103Brain Tumor Detection and ClassificationAI in cancer detectionGlioma Diagnosis and Treatment

A Comprehensive Analysis of Recent Deep and Federated-Learning-Based Methodologies for Brain Tumor Diagnosis

Open access
Naeem, Ahmad, Anees, Tayyaba, Naqvi, Rizwan Ali, Loh, Woong-Kee·2022·Journal of Personalized Medicine
DOI 10.3390/jpm12020275PMC PMC8880689
Open this paper in CoChat — free accountOpen PDF

Abstract

Brain tumors are a deadly disease with a high mortality rate. Early diagnosis of brain tumors improves treatment, which results in a better survival rate for patients. Artificial intelligence (AI) has recently emerged as an assistive technology for the early diagnosis of tumors, and AI is the primary focus of researchers in the diagnosis of brain tumors. This study provides an overview of recent research on the diagnosis of brain tumors using federated and deep learning methods. The primary objective is to explore the performance of deep and federated learning methods and evaluate their accuracy in the diagnosis process. A systematic literature review is provided, discussing the open issues and challenges, which are likely to guide future researchers working in the field of brain tumor diagnosis.

AI snapshot

Written from the paper’s full text.

TL;DR

This systematic literature review surveys 55 primary studies on deep-learning- and federated-learning-based brain tumor diagnosis, proposing a taxonomy and a common diagnosis model. It reports that ensemble and handcrafted classifiers perform strongly, BraTS and Figshare are widely used, and federated learning addresses limited datasets without sharing patient data.

Key findings

  • From 3986 initially identified studies, screening yielded 55 primary studies: 221 after title screening, 88 after abstract review, 52 after full-text review, plus 3 added through snowball tracking.
  • The review concludes that ensemble and handcrafted classifiers show outstanding results for brain tumor detection; a 75:25 training/testing ratio provided the best accuracy in the literature, and the BraTS challenge and Figshare (CE-MRI) datasets were most widely used.
  • Federated learning is presented as resolving limited training/testing datasets without compromising data privacy; selected federated learning works include Sheller et al., Li et al., and Guo et al. for segmentation or MR image reconstruction.
  • Open issues and challenges include dataset variations, varying numbers of low-grade and high-grade tumor images, difficulty detecting lesions smaller than 3 mm, and data collected from patients with an average age between 40 and 70 years, with no data from younger patients.

Methods

  • Systematic literature review following a defined search and screening process; repositories searched included Elsevier, ACM, IEEE, Springer, MDPI, Wiley, Miccai, and Medline from 1 January 2017 to 20 December 2021, with other digital libraries excluded due to accessibility constraints.
  • Inclusion/exclusion criteria excluded research not based on binary disease classification, diagnosis without medical images, studies with unclear data sources, and studies based on non-human samples.
  • The 55 selected papers were categorized into pretrained, handcrafted, ensemble, and federated learning methods; performance metrics discussed include sensitivity, specificity, precision, accuracy, and area under the curve (AUC), and a taxonomy and common diagnosis model were proposed.

Limitations

  • The paper states that other digital libraries were not included due to accessibility constraints, and that the diversity of available datasets makes it difficult to compare and validate results; increasing precision remains an obvious challenge.
  • The paper does not explicitly state limitations of its own review; evident from the text, it is a qualitative systematic review without pooled quantitative meta-analysis, and many included studies are criticized for single-dataset or small-sample validation.
  • Also evident from the text, the search was restricted to 2017–2021 and excluded non-binary classification, non-image, and non-human studies, which may narrow the scope of the review.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/jpm12020275

Cites

Statistics

Nothing to recompute.

Claims the paper makes

0 of 77 cited sources checked · 0 supported · 0 flagged

  • Findingcitation

    The use of two datasets enhanced the accuracy of the proposed method by Sultan et al.

    “The use of the two datasets enhanced the accuracy of the proposed method.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [42] Sultan, 2019, Multi-Classification of Brain Tumor Images Using Deep Neural Network, IEEE Access source

      Not checked in the free run.

  • Findingreasoning

    The main shortcoming of pre-trained methods is that the efficiency of the proposed methods was only measured on a single dataset.

    “The main shortcoming of these pre-trained methods is that the efficiency of the proposed methods was only measured on a single dataset.”
    Quote in the paperNot a substantial claim
  • Findingcitation

    The proposed method by Das et al. used only one dataset to measure effectiveness.

    “The main limitation is that only one dataset was used to measure the effectiveness of the proposed method.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] Das, S., Aranya, O.R.R., and Labiba, N.N. (2019, January 3–5). Brain Tumor Classification Using Convolutional Neural Network. Proceedings of the 2019 1st International Conference on Advances in Science, Engineering and Robotics Technology (ICASERT), Dhaka, Bangladesh. source

      Not checked in the free run.

  • Findingcitation

    Different image dimensions are used in different datasets.

    “Different image dimensions are used in different datasets [40].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [40] Guo, P., Wang, P., Zhou, J., Jiang, S., and Patel, V.M. (2021, January 20–25). Multi-institutional collaborations for improving deep learning-based magnetic resonance image reconstruction using federated learning. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2021, Nashville, TN, USA. source

      Not checked in the free run.

  • Findingcitation

    In the literature, all data are collected from patients with an average age between 40 and 70 years.

    “It was observed that in the literature, all the data are collected from patients with an average age between 40 and 70 years.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [22] (2021, December 10). Centre for Reviews and Dissemination | Home. Available online: http://www.york.ac.uk/inst/crd/faq4.htm.

      Not checked in the free run.

  • Classified as probably not a substantial claim. 65
  • May combine two claims. 28

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    After abstract review, 88 papers remained on the list.

    “After that stage, 88 papers remained on the list.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    After full-text review, 52 papers were considered for analysis.

    “At this stage, 52 papers were considered for the analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In the first phase, 3986 studies were identified using the search strings.

    “In the first phase, 3986 studies were identified using the search strings.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    After title screening, 221 articles were classified as appropriate.

    “At this stage, 221 articles were classified as appropriate.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Snowball tracking added three articles, resulting in a total of 55 primary studies.

    “three more articles were thus included in the list of selected papers, resulting in a total of 55 primary studies.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The 55 selected papers were collected from journals, conferences, and symposiums.

    “The 55 selected papers were collected from different publication channels, which include journals, conferences, and symposiums.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    The proposed method by Sair et al. was tested on a single dataset, which lowers the credibility of the method.

    “The proposed method was tested on a single dataset, which lowers the credibility of the method.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [41] Siar, M., and Teshnehlab, M. (2019, January 24–25). Brain Tumor Detection Using Deep Neural Network and Machine Learning Algorithm. Proceedings of the 2019 9th International Conference on Computer and Knowledge Engineering (ICCKE), Mashhad, Iran. source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    The images in the brain MRNet dataset were not of uniform resolution and image quality was not high, reducing the efficiency of the proposed method.

    “The images were not of a uniform resolution, and the image quality was not high, reducing the efficiency of the proposed method.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [59] Ergen, 2020, BrainMRNet: Brain tumor detection using magnetic resonance images with a novel convolutional neural network model, Med. Hypotheses source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    These datasets vary in the number of images.

    “These datasets vary in the number of images [55].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [55] Sharif, M.I., Khan, M.A., Alhussein, M., Aurangzeb, K., and Raza, M. (2021). A decision support system for multimodal brain tumor classification using deep learning. Complex Intell. Syst. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    Gender, age, and race were added to achieve better results.

    “Furthermore, gender, age, and race were added to achieve better results.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    Every dataset has a varying number of low-grade and high-grade tumors.

    “Every dataset has a varying number of low-grade and high-grade tumors.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    Some researchers created a dataset by collecting images from patients in different hospitals.

    “Some researchers created a dataset by collecting images from patients in different hospitals.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Findingnone given

    Ensemble and handcrafted classifiers show outstanding results for brain tumor detection.

    “It has been observed from the selected literature that ensemble and handcrafted classifiers show outstanding results for brain tumor detection.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    Some studies have used handcrafted features to perform pre-processing for the segmentation and extraction of features.

    “Some studies have also used handcrafted features to perform pre-processing for the segmentation and extraction of features.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    The BraTS challenge and Figshare (CE-MRI) datasets have been widely used by researchers.

    “It has been observed that in the literature, the BraTS challenge and Figshare (CE-MRI) datasets have been widely used by researchers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    Researchers have also used non-public and self-collected datasets for the diagnosis of brain tumors.

    “Researchers have also used non-public and self-collected datasets for the diagnosis of brain tumors.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    The performance of ensemble classifiers works best on large datasets.

    “The performance of ensemble classifiers works best on large datasets, more data validates the accuracy of the classifier.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Multiple datasets are needed to validate the effectiveness or reliability of brain tumor diagnosis methods.

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The search retrieved articles from Elsevier, ACM, IEEE, Springer, MDPI, Wiley, Miccai, and Medline from 1 January 2017 to 20 December 2021.

    “Elsevier, ACM, IEEE, Springer, MDPI, Wiley, Miccai, and Medline were searched from 1 January 2017, to 20 December 2021, to retrieve relevant articles.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The Figshare dataset has been publicly available since 2015.

    “This dataset has been publicly available since 2015.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [42] Sultan, 2019, Multi-Classification of Brain Tumor Images Using Deep Neural Network, IEEE Access source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    The latest version of the dataset released in 2017 is commonly known as the Figshare dataset.

    “The latest version released in 2017 is commonly known as the Figshare dataset.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The Figshare dataset contains 3064 T1-weighted contrast-enhanced images collected from 233 patients.

    “a publicly available dataset from Figshare containing 3064 T1-weighted contrast-enhanced images collected from 233 patients.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] Das, S., Aranya, O.R.R., and Labiba, N.N. (2019, January 3–5). Brain Tumor Classification Using Convolutional Neural Network. Proceedings of the 2019 1st International Conference on Advances in Science, Engineering and Robotics Technology (ICASERT), Dhaka, Bangladesh. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The brain MRNet database consisted of 253 publicly available MR images labeled normal and tumor, with 155 malignant and 98 normal tumor images.

    “The total number of images was 253, and patients provided each image voluntarily, making the dataset heterogeneous.”
    • Also stated in: § 3.4. RQ2 Assessment: What Are the Metrics Used to Determine the Performance of Different Methods Applied to Brain Tumor Diagnosis? · plain
      “There were 155 malignant tumors and 98 normal tumor images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [59] Ergen, 2020, BrainMRNet: Brain tumor detection using magnetic resonance images with a novel convolutional neural network model, Med. Hypotheses source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The Figshare dataset consists of 708 meningiomas, 1426 gliomas, and 930 pituitary images.

    “The data consisted of three different kinds of brain tumors: 708 meningiomas, 1426 gliomas, and 930 pituitary images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] Das, S., Aranya, O.R.R., and Labiba, N.N. (2019, January 3–5). Brain Tumor Classification Using Convolutional Neural Network. Proceedings of the 2019 1st International Conference on Advances in Science, Engineering and Robotics Technology (ICASERT), Dhaka, Bangladesh. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The BraTS 2017 dataset consists of multimodal MRIs from various institutions.

    “The BraTS 2017 dataset consists of multimodal MRIs from various institutions.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [78] (2021, December 10). BraTS Challenge | Start 2017. Available online: https://www.med.upenn.edu/sbia/brats2017/data.html.

      Not checked in the free run.

    • cites [79] (2021, December 10). BraTS Challenge | Start 2016. Available online: https://www.smir.ch/BraTS/Start2016.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    BraTS 2014 has data from 300 input subjects, of which 200 were used for training of HGG cases and 100 were used for testing of LGG cases.

    “This dataset has data from 300 input subjects, of which 200 were used for the training of HGG cases and 100 were used for the testing of LGG cases [76].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [76] (2021, December 10). BraTS Challenge | Start 2014. Available online: https://www.virtualskeleton.ch/BRATS/Start2014.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Snowball tracking was carried out by searching the references of selected studies to ensure that no important research was missing.

    “Moreover, snowball tracking was carried out by searching the references of selected studies to ensure that no important research was missing.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Other digital libraries were explored but were not included due to accessibility constraints.

    “Other digital libraries were also explored but were not included due to accessibility constraints.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The BraTS 2017 dataset was designed to segment inherently heterogeneous brain tumors.

    “This dataset was designed to segment inherently heterogeneous brain tumors.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [78] (2021, December 10). BraTS Challenge | Start 2017. Available online: https://www.med.upenn.edu/sbia/brats2017/data.html.

      Not checked in the free run.

    • cites [79] (2021, December 10). BraTS Challenge | Start 2016. Available online: https://www.smir.ch/BraTS/Start2016.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The defined search process by Dybå et al. was adopted for screening relevant papers.

    “For the screening of relevant papers, the defined search process by Dybå et al. [19] was adopted.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] 2008, Empirical studies of agile software development: A systematic review, Inf. Softw. Technol. source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductcitation

    Articles were excluded based on criteria: not based on binary disease classification; diagnosing brain tumors without medical images; not identifying data sources or employing ambiguous methods of data collection; based on non-human samples.

    “Articles were excluded based on the following criteria [21]: Research articles not based on binary disease classification.Research articles diagnosing brain tumors without medical images.Research articles not identifying data sources or employing ambiguous methods of data collection.Research articles based on non-human samples.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [21] Malik, 2020, A Comparison of Transfer Learning Performance Versus Health Experts in Disease Diagnosis from Medical Imaging, IEEE Access source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    TCGA-GBM consists of 500 different samples of brain cancer.

    “The cancer genome atlas glioblastoma multiforme (TCGA-GBM) consists of 500 different samples of brain cancer.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [73] Scarpace, 2016, Radiology data from the cancer genome atlas glioblastoma multiforme [TCGA-GBM] collection, Cancer Imaging Arch.

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    BraTS 2015 had a total of 284 subjects, of which 64 belonged to LGG and 220 belonged to HGG cases; 174 subjects were used for training and 110 subjects were used for testing.

    “In BraTS 2015, there were a total of 284 subjects, of which 64 belonged to LGG and 220 belonged to HGG cases; of the total, 174 subjects were used for training and 110 subjects were used for testing [77].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [77] (2021, December 10). BraTS Challenge | Start 2015. Available online: https://www.smir.ch/BraTS/Start2015.

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Few studies have reviewed deep and federated learning, and these studies have limitations.

    “Few studies have reviewed deep and federated learning. Moreover, these studies have limitations.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [11] Nalepa, 2019, Data Augmentation for Brain-Tumor Segmentation: A Review, Front. Comput. Neurosci. source

      Not checked in the free run.

    • cites [14] Awad, 2019, A review on brain tumor diagnosis from MRI images: Practical implications, key achievements, and lessons learned, Magn. Reson. Imaging source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Federated learning is used to train the shared global model using data from several institutions without compromising data privacy.

    “To overcome this issue, federated learning is used to train the shared global model using data from several institutions without compromising data privacy [13].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [13] Xu, 2021, Federated Learning for Healthcare Informatics, J. Health Inform. Res. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Progress in deep learning has contributed to many state-of-the-art brain tumor identification solutions, allowing early cancer discovery.

    “Progress in deep learning, which is a combination of AI and machine learning, has contributed to many state-of-the-art brain tumor identification solutions, allowing early cancer discovery, whereby preemptive measures can be taken to save lives [12].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] Ritchie, H. (2021, December 10). How Many People in the World Die from Cancer| Our World in Data| Institute for Health Metrics and Evaluation (IHME). Available online: https://ourworldindata.org/how-many-people-in-the-world-die-from-cancer.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The Figshare dataset contains 3064 T1-weighted contrast-enhanced images collected from 233 patients with a resolution of 512 × 512 pixels.

    “This contains 3064 T1-weighted contrast-enhanced images collected from 233 patients with a resolution of 512 × 512 pixels.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [72] Cheng, J., Huang, W., Cao, S., Yang, R., Yang, W., Yun, Z., Wang, Z., and Feng, Q. (2015). Enhanced performance of brain tumor classification via tumor region augmentation and partition. PLoS ONE, 10. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Both BraTS 2012 and 2013 have a total of 30 HGG/LGG patients' data, including 20 HGG and 10 LGG cases.

    “Both BraTS 2012 and 2013 have a total of 30 HGG/LGG patients’ data, which includes 20 HGG and 10 LGG cases [74,75].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [74] (2021, December 10). Cham.BraTS Challenge | Start 2012. Available online: https://www.smir.ch/BRATS/Start2012.

      Not checked in the free run.

    • cites [75] (2021, December 10). BraTS Challenge | Start 2013. Available online: https://www.smir.ch/BRATS/Start2013.

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Abd-Ellah et al. performed a systematic literature review on brain tumor detection, segmentation, and classification using machine and deep learning, but did not provide an overview of recent deep learning methods and the BraTS 2018 dataset.

    “Another systematic literature review was performed by Abd-Ellah et al. [14] on brain tumor detection, segmentation, and classification using machine and deep learning, but they did not provide an overview of recent deep learning methods and the BraTS 2018 dataset.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [14] Awad, 2019, A review on brain tumor diagnosis from MRI images: Practical implications, key achievements, and lessons learned, Magn. Reson. Imaging source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    In experimental studies using TCGA-GBM, T1-weighted post-contrast (T1-gadolinium) sequence images were used to obtain the most realistic results in the tumor region of the brain MRI.

    “In experimental studies, the images in the T1-weighted post-contrast (T1-gadolinium (Gd)) sequence were used to obtain the most realistic results in the tumor region of the brain MRI [73].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [73] Scarpace, 2016, Radiology data from the cancer genome atlas glioblastoma multiforme [TCGA-GBM] collection, Cancer Imaging Arch.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisereasoning

    TCGA-GBM is an open-access dataset that does not require approval of an ethics committee.

    “It is an open-access dataset provided by the TCGA-GBM for researchers to conduct scientific studies on brain tumors. Therefore, this dataset does not require the approval of an ethics committee.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [73] Scarpace, 2016, Radiology data from the cancer genome atlas glioblastoma multiforme [TCGA-GBM] collection, Cancer Imaging Arch.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    AI-based methods are used for brain tumor detection because of their outstanding results.

    “AI-based methods are used for brain tumor detection because of their outstanding results [11].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [11] Nalepa, 2019, Data Augmentation for Brain-Tumor Segmentation: A Review, Front. Comput. Neurosci. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Primary tumors are categorized as benign or malignant depending on their type, and they can be life-threatening because when they grow, they cause damage to brain cells and skulls.

    “These tumors are categorized as benign or malignant, depending on their type, and they can be life-threatening because when they grow, they cause damage to brain cells and skulls [14].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] Awad, 2019, A review on brain tumor diagnosis from MRI images: Practical implications, key achievements, and lessons learned, Magn. Reson. Imaging source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The BraTS 2017 training package included 285 samples with manually annotated and confirmed ground values.

    “The training package included 285 samples with manually annotated and confirmed ground values.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [78] (2021, December 10). BraTS Challenge | Start 2017. Available online: https://www.med.upenn.edu/sbia/brats2017/data.html.

      Not checked in the free run.

    • cites [79] (2021, December 10). BraTS Challenge | Start 2016. Available online: https://www.smir.ch/BraTS/Start2016.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Secondary tumors of the brain are malignant.

    “Secondary tumors of the brain are malignant.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] Cancer Research UK (2022, January 30). Brain Tumor. Available online: https://www.cancerresearchuk.org/about-cancer/brain-tumours.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Two authors carried out the process of quality assessment for the selected studies.

    “Two authors carried out the process of quality assessment for the selected studies.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    BraTS 2017 provided four modalities (T1, T1c, T2, and FLAIR) for each sample, along with corresponding annotations; the image size was 240 × 240 × 155 pixels.

    “Four modalities, including T1, T1c, T2, and FLAIR were provided for each sample, along with corresponding annotations; the image size was 240 × 240 × 155 pixels [78,79].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [78] (2021, December 10). BraTS Challenge | Start 2017. Available online: https://www.med.upenn.edu/sbia/brats2017/data.html.

      Not checked in the free run.

    • cites [79] (2021, December 10). BraTS Challenge | Start 2016. Available online: https://www.smir.ch/BraTS/Start2016.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Other primary tumor types include pituitary tumors, pineal gland tumors, ependymomas, craniopharyngiomas, lymphomas, germ cell tumors, and schwannomas.

    “The other types include pituitary tumors, pineal gland tumors, ependymomas, craniopharyngiomas, lymphomas, germ cell tumors, and schwannomas [87].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [87] Ostrom, 2019, Risk factors for childhood and adult primary brain tumors, Neuro-Oncology source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    BraTS 2019 uses preoperative MRI scans and mainly focuses on gliomas.

    “Preoperative MRI scans are used in BraTS 2019, mainly focusing on gliomas (intrinsically complex brain tumors).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [84] (2021, December 10). BraTS Challenge | Start 2019. Available online: https://www.med.upenn.edu/cbica/brats-2019.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Publicly available benchmark datasets Figshare, BraTS challenge 2013, 2014, 2015, 2017, and 2018, ISLES 2015 and 2017, TCGA-GBM, and Brain MRI are accessible to all.

    “The publicly available benchmark datasets, Figshare, BraTS challenge 2013, 2014, 2015, 2017, and 2018, ISLES 2015 and 2017, TCGA-GBM, and Brain MRI are accessible to all.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    BraTS 2018 had 285 input subjects, of which 75 belonged to LGG and 210 belonged to HGG; 191 cases were used for testing.

    “BraTS 2018 also had 285 input subjects, of which 75 belonged to LGG and 210 belonged to HGG; 191 cases were used for testing.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [80] Zhao, 2018, A deep learning model integrating FCNNs and CRFs for brain tumor segmentation, Med. Image Anal. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationreasoning

    The model for the detection of brain tumors helps researchers overcome existing issues in current research.

    “The model for the detection of brain tumors helps researchers overcome existing issues in current research”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The ISLES 2017 dataset has 75 input subjects; 28 subjects were used for training and 36 for testing.

    “This dataset has 75 input subjects; 28 subjects were used for training and 36 for testing [82].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [82] (2021, December 10). Ischemic Stroke Lesion Segmentation | ISLES 2017. Available online: http://www.isles-challenge.org/ISLES2017/.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The BraTS 2019 dataset also predicts the overall survival of patients.

    “The dataset also predicts the overall survival of patients [84].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [84] (2021, December 10). BraTS Challenge | Start 2019. Available online: https://www.med.upenn.edu/cbica/brats-2019.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Secondary brain tumors initially develop in one part of the body and can spread to the brain; they consist of lung cancer, breast cancer, kidney cancer, and skin cancer.

    “Tumors initially develop in one part of the body and can spread to the brain. These tumors consist of lung cancer, breast cancer, kidney cancer, and skin cancer.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] Cancer Research UK (2022, January 30). Brain Tumor. Available online: https://www.cancerresearchuk.org/about-cancer/brain-tumours.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Benign tumors are not distributed from one body part to another.

    “Benign tumors are not distributed from one body part to another [1].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] Cancer Research UK (2022, January 30). Brain Tumor. Available online: https://www.cancerresearchuk.org/about-cancer/brain-tumours.

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Classification of abnormal MR images into different tumor categories is still a challenging task.

    “However, the classification of these abnormal images into different tumor categories is still a challenging task [21].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [21] Malik, 2020, A Comparison of Transfer Learning Performance Versus Health Experts in Disease Diagnosis from Medical Imaging, IEEE Access source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The ISLES 2015 dataset has 64 subjects with subacute stroke lesion segmentation (SISS); it provided 42 subjects for training and 22 for testing.

    “This dataset has 64 subjects with subacute stroke lesion segmentation (SISS). This dataset provided 42 subjects for training and 22 for testing [81].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [81] (2021, December 10). Ischemic Stroke Lesion Segmentation | ISLES 2015. Available online: http://www.isles-challenge.org/ISLES2015/.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The BRAINIX dataset is only available on special request.

    “This data is only available on special request [85].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [85] (2021, December 10). DICOM Image Library | Home. Available online: https://www.osirix-viewer.com/resources/dicom-image-library/brainix.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The Combined dataset consists of 15,320 MRI images collected from different sources; the data are accessible upon request only under a license agreement.

    “This dataset consists of 15,320 MRI images collected from different sources. The data are accessible upon request only under a license agreement [57].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [57] Gumaei, 2019, A Hybrid Feature Extraction Method with Regularized Extreme Learning Machine for Brain Tumor Classification, IEEE Access source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Gliomas and meningiomas are the most common form of tumors and originate from the cerebrum and supportive tissues of the brain.

    “Gliomas and meningiomas are the most common form of tumors [43] and originate from the cerebrum and supportive tissues of the brain [42,86].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [43] Li, 2019, Brain Tumor Detection Based on Multimodal Information Fusion and Convolutional Neural Network, IEEE Access source

      Not checked in the free run.

    • cites [42] Sultan, 2019, Multi-Classification of Brain Tumor Images Using Deep Neural Network, IEEE Access source

      Not checked in the free run.

    • cites [86] Guadagno, E., Presta, I., Maisano, D., Donato, A., Pirrone, C.K., Cardillo, G., Corrado, S.D., Mignogna, C., Mancuso, T., and Donato, G. (2018). Role of Macrophages in Brain Tumor Growth and Progression. Int. J. Mol. Sci., 19. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The Brain MRI dataset has 253 input subjects collected from volunteer patients; the study included 155 tumor subjects and 98 normal subjects.

    “This dataset has 253 input subjects that were collected from volunteer patients. The study included 155 tumor subjects and 98 normal subjects [83].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [83] Kaggle (2021, December 10). Navoneel | Brain Tumor. Available online: https://www.kaggle.com/navoneel/brain-mri-images-for-brain-tumor-detection.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The model for brain tumor detection was finalized after analysis of best-performing methods existing in the literature.

    “The model was finalized after the analysis of best-performing methods existing in the literature, such as [43,52,53,58,66].”
    Quote in the paperNot a substantial claim0 of 5 sources checked
    • cites [43] Li, 2019, Brain Tumor Detection Based on Multimodal Information Fusion and Convolutional Neural Network, IEEE Access source

      Not checked in the free run.

    • cites [52] Saba, 2020, Brain tumor detection using fusion of hand crafted and deep learning features, Cogn. Syst. Res. source

      Not checked in the free run.

    • cites [53] Amin, 2020, Brain Tumor Detection by Using Stacked Autoencoders in Deep Learning, J. Med. Syst. source

      Not checked in the free run.

    • cites [58] Amin, 2018, Big data analysis for brain tumor detection: Deep convolutional neural networks, Future Gener. Comput. Syst. source

      Not checked in the free run.

    • cites [66] Amin, 2019, A new approach for brain tumor segmentation and classification based on score level fusion using transfer learning, J. Med. Syst. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    In the non-public category, a combined dataset is available with images collected from different sources, and Brainix is available upon special request for brain tumor diagnosis.

    “In the non-public category, a combined dataset is available with images collected from different sources. Brainix is also available upon special request for brain tumor diagnosis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    These tumors can cause vomiting, blurred vision, dizziness, memory loss, loss of balance, slow mental response, confusion, and difficulty in writing and reading.

    “These tumors can cause vomiting, blurred vision, dizziness, memory loss, loss of balance, slow mental response, confusion, and difficulty in writing and reading [7].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [7] (2021, December 10). Brain Tumor Informatics| Home. Available online: https://braintumor.org/brain-tumor-information.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    A questionnaire was formed to evaluate the quality of selected papers.

    “A questionnaire was formed which evaluates the quality of selected papers [18].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] Naeem, 2020, Malignant Melanoma Classification Using Deep Learning: Datasets, Performance Measurements, Challenges and Opportunities, IEEE Access source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    BraTS 2018 image dimensions were 240 × 240 × 155 pixels.

    “The dimensions of the images were 240 × 240 × 155 pixels [80].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [80] Zhao, 2018, A deep learning model integrating FCNNs and CRFs for brain tumor segmentation, Med. Image Anal. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    The issue of the limited dataset for training and testing is resolved by federated learning without compromising the privacy of data.

    “Moreover, the issue of the limited dataset for training and testing is resolved by federated learning without compromising the privacy of data.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    With deep learning techniques, there is no dire need for composite and complex pre-processing techniques, such as image resizing, cropping, and normalization of the pixel values.

    “With deep learning techniques, there is no dire need for composite and complex pre-processing techniques, such as image resizing, cropping, and normalization of the pixel values.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Shakeel et al. utilized infrared imaging technology to detect brain tumors of less than 3 mm.

    “To address this challenge, Shakeel et al. [46] utilized infrared imaging technology to detect brain tumors of less than 3 mm.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [46] Shakeel, 2019, Neural Network Based Brain Tumor Detection Using Wireless Infrared Imaging Sensor, IEEE Access source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    A variety of benchmark datasets, such as the BraTS challenge 2013, 2014, 2015, 2017, 2018, and 2019 datasets, are available for researchers in this domain to validate their work on publicly available datasets.

    “A variety of benchmark datasets, such as the BraTS challenge 2013, 2014, 2015, 2017, 2018, and 2019 datasets, are available for researchers in this domain to validate their work on publicly available datasets.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Increasing precision remains an obvious challenge.

    “However, increasing the precision remains an obvious challenge.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Compared to conventional methods, deep-learning-based methods produce better results.

    “Compared to conventional methods, deep-learning-based methods produce better results.”
    Quote in the paperNot a substantial claim
72 of 72

  • The paper relies on this without support.

    In recent years, much work has been performed on the automated diagnosis of brain tumors using deep learning, whereas only a few studies have been conducted on federated learning.

    “In recent years, much work has been performed on the automated diagnosis of brain tumors using deep learning, whereas only a few studies have been conducted on federated learning.”

  • The paper relies on this without support.

    Clinicians still depend on manual tumor projections, probably due to a lack of communication between clinicians and researchers.

    “Despite a reasonable amount of work done in this field, however, clinicians still depend on manual tumor projections. This is probably due to a lack of communication between clinicians and researchers.”

  • The paper relies on this without support.

    Multiple datasets are needed to validate the effectiveness or reliability of brain tumor diagnosis methods.

    “Multiple datasets must be used to measure the effectiveness of the proposed method.”

  • The paper relies on this without support.

    Publicly available datasets are necessary to validate the performance of proposed brain tumor diagnosis methods.

    “A publicly available dataset must be used to validate the performance of the proposed method.”

  • The paper relies on this without support.

    The selected 55 primary studies are representative of the state-of-the-art in deep and federated learning for brain tumor diagnosis.

    “The 55 selected papers were collected from different publication channels, which include journals, conferences, and symposiums.”

  • The authors state this as an assumption.

    Publishing research in Q1, Q2, Q3, and Q4 journals is more difficult than publishing in other publication venues.

    “since the authors believe that publishing research in Q1, Q2, Q3, and Q4 could be more difficult as compared to other publications”

  • The paper relies on this without support.

    Brain tumors have a similar appearance and structure, making manual detection of brain tumors using MR images very difficult and time consuming for the radiologist.

    “Brain tumors have a similar appearance and structure, and due to this, the manual detection of brain tumors using MR images is very difficult and time consuming for the radiologist.”

  • The paper relies on this without support.

    Expert radiologists can easily differentiate between normal and abnormal MR images.

    “The expert radiologist can easily differentiate between normal and abnormal MR images.”

  • The paper relies on this without support.

    Brain tumors occur in irregular shapes, making it difficult for the radiologist to manually detect a specific type of tumor from MR images.

    “Brain tumors occur in irregular shapes, and it is a very difficult task for the radiologist to manually detect a specific type of tumor from MR images.”

  • The paper relies on this without support.

    Grade I meningioma is benign, whereas grade II and grade III have a high risk of recurrence, invasiveness, and aggressiveness.

    “Grade I is considered benign, whereas grade II and Grade III have a high risk of recurrence, invasiveness, and aggressiveness.”

  • The paper relies on this without support.

    It is difficult to identify a brain tumor if the size of the lesion is less than 3 mm.

    “It is difficult to identify the tumor if the size of the lesion is less than 3 mm.”

  • The paper relies on this without support.

    Primary tumors develop because of the abnormal growth of cells in the brain.

    “Primary tumors develop because of the abnormal growth of cells in the brain.”

  • The paper relies on this without support.

    Primary tumors can develop from brain membranes, brain cells, nerve cells, and glands.

    “Primary tumors have many types, and they can develop from brain membranes, brain cells, nerve cells, and glands.”

  • The paper relies on this without support.

    Some brain tumors are cancerous, while others are non-cancerous.

    “Some of these tumors are cancerous, while others are non-cancerous.”

  • The paper relies on this without support.

    The majority of brain cancers consist of secondary brain tumors.

    “The majority of brain cancers consist of secondary brain tumors.”
15 of 15

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.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Binary-only screening conflicts with included multiclassification studies

    Major

    The screening criteria exclude articles not based on binary disease classification, but the paper describes an included method that classifies brain tumors into three types.

  • Claim of no younger-patient data conflicts with reported age range

    Minor

    The paper says no data were collected from younger patients, but it also describes a study using patient data spanning ages 6 to 80.

  • Figshare patient counts disagree

    Minor

    The paper identifies a dataset of images from 130 patients as Figshare, but later describes the Figshare dataset as containing images from 233 patients.

3 of 3

The citation lists haven’t been fetched yet

References 70 · Cited by 103

Research works printed but not linked: 1

  1. Scarpace, 2016, Radiology data from the cancer genome atlas glioblastoma multiforme [TCGA-GBM] collection, Cancer Imaging Arch., 11, 1

Authors
Naeem, A., Anees, T., Naqvi, R. A., & Loh, W.-K.
Affiliations
  • University of Management and Technology (PK)
  • Sejong University (KR)
  • Gachon University (KR)
Corresponding author
  • Naqvi, Rizwan Ali
  • Loh, Woong-Kee
ORCID
Venue
Journal of Personalized Medicine
Publisher
MDPI AG
Published
2022-02-13
Volume
12
Issue
2
Pages
275
ISSN
2075-4426 (ISSN-L)
DOI
10.3390/jpm12020275
PMC
PMC8880689
OpenAlex
W4212858816
Cited by
103
References
70
Citations by year
2026: 13 · 2025: 31 · 2024: 27 · 2023: 30 · 2022: 2
FWCI
2.32
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Personalized Medicine (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Life Sciences › Neuroscience › Neurology › Brain Tumor Detection and Classification
  • Physical Sciences › Computer Science › Artificial Intelligence › AI in cancer detection
  • Health Sciences › Medicine › Genetics › Glioma Diagnosis and Treatment
Keywords
brain tumor diagnosis, federated learning, deep learning, early diagnosis, artificial intelligence in healthcare, diagnostic accuracy, medical image analysis

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 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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