Dataset Information

Improving model fairness in image-based computer-aided diagnosis.

ABSTRACT: Deep learning has become a popular tool for computer-aided diagnosis using medical images, sometimes matching or exceeding the performance of clinicians. However, these models can also reflect and amplify human bias, potentially resulting inaccurate missed diagnoses. Despite this concern, the problem of improving model fairness in medical image classification by deep learning has yet to be fully studied. To address this issue, we propose an algorithm that leverages the marginal pairwise equal opportunity to reduce bias in medical image classification. Our evaluations across four tasks using four independent large-scale cohorts demonstrate that our proposed algorithm not only improves fairness in individual and intersectional subgroups but also maintains overall performance. Specifically, the relative change in pairwise fairness difference between our proposed model and the baseline model was reduced by over 35%, while the relative change in AUC value was typically within 1%. By reducing the bias generated by deep learning models, our proposed approach can potentially alleviate concerns about the fairness and reliability of image-based computer-aided diagnosis.

SUBMITTER: Lin M

PROVIDER: S-EPMC10558498 | biostudies-literature | 2023 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Improving model fairness in image-based computer-aided diagnosis.

Lin Mingquan M Li Tianhao T Yang Yifan Y Holste Gregory G Ding Ying Y Van Tassel Sarah H SH Kovacs Kyle K Shih George G Wang Zhangyang Z Lu Zhiyong Z Wang Fei F Peng Yifan Y

Nature communications 20231006 1

Deep learning has become a popular tool for computer-aided diagnosis using medical images, sometimes matching or exceeding the performance of clinicians. However, these models can also reflect and amplify human bias, potentially resulting inaccurate missed diagnoses. Despite this concern, the problem of improving model fairness in medical image classification by deep learning has yet to be fully studied. To address this issue, we propose an algorithm that leverages the marginal pairwise equal op ...[more]

PMID: 37803009

Similar Datasets

Project description:BackgroundAlthough miniscrews are widely used in orthodontic treatment as temporary anchorage devices, their correct and safe placement has attracted little attention. This study aimed to introduce a novel cone beam CT (CBCT) image-based computer aided design and computer aided manufacturing (CAD-CAM) template for orthodontic miniscrew implantation and to evaluate the effectiveness of miniscrews implanted under the guidance of this template.MethodsThe CBCT scans of ten patients requiring miniscrews as anchorage were analyzed in NNT software to predetermine the insertion sites of miniscrew implants. The DICOM data of the scans, along with virtual miniscrews acquired in Solidworks software, were imported into Mimics software to construct three-dimensional (3D) images of teeth and bone and to determine the virtual position of miniscrews and 3D virtual templates were designed following consideration of the virtual implantation plans. A STL (Stereolithography) file of the virtual template was output, and the resin template was then fabricated with a stereolithographic appliance (SLA). 24 Miniscrews were then implanted guided by the template and clinical evaluation of their safety and stability, as well as their placement deviations, were made. A dental casts model and cephalometric analysis before and after orthodontic treatment were made to assess the dentomaxillofacial changes.ResultsAll 24 miniscrews had no contact with adjacent roots. 18 miniscrews had a grade I safety score and six had a grade II. The miniscrews were stable at 1, 3, 6, and 9 months after implantation, although there was mild inflammation around two miniscrews. Implantation deviation of miniscrew in the crown was (1.03±0.65) and (1.26±0.72) mm in the apex, on average. Satisfactory dentomaxillofacial changes in 10 patients with these 24 miniscrews as anchorage were acquired.ConclusionsMiniscrews could be implanted in the targeted position safely and precisely when guided by the novel templates, and remained stable during orthodontic treatment. Patients treated with these miniscrews as anchorage in orthodontic treatment acquired satisfactory dentomaxillofacial changes.

Project description:Training deep Convolutional Neural Networks (CNNs) presents challenges in terms of memory requirements and computational resources, often resulting in issues such as model overfitting and lack of generalization. These challenges can only be mitigated by using an excessive number of training images. However, medical image datasets commonly suffer from data scarcity due to the complexities involved in their acquisition, preparation, and curation. To address this issue, we propose a compact and hybrid machine learning architecture based on the Morphological and Convolutional Neural Network (MCNN), followed by a Random Forest classifier. Unlike deep CNN architectures, the MCNN was specifically designed to achieve effective performance with medical image datasets limited to a few hundred samples. It incorporates various morphological operations into a single layer and uses independent neural networks to extract information from each signal channel. The final classification is obtained by utilizing a Random Forest classifier on the outputs of the last neural network layer. We compare the classification performance of our proposed method with three popular deep CNN architectures (ResNet-18, ShuffleNet-V2, and MobileNet-V2) using two training approaches: full training and transfer learning. The evaluation was conducted on two distinct medical image datasets: the ISIC dataset for melanoma classification and the ORIGA dataset for glaucoma classification. Results demonstrate that the MCNN method exhibits reliable performance in melanoma classification, achieving an AUC of 0.94 (95% CI: 0.91 to 0.97), outperforming the popular CNN architectures. For the glaucoma dataset, the MCNN achieved an AUC of 0.65 (95% CI: 0.53 to 0.74), which was similar to the performance of the popular CNN architectures. This study contributes to the understanding of mathematical morphology in shallow neural networks for medical image classification and highlights the potential of hybrid architectures in effectively learning from medical image datasets that are limited by a small number of case samples.

Project description:IntroductionAnatomic characteristics of kidneys derived from ultrasound images are potential biomarkers of children with congenital abnormalities of the kidney and urinary tract (CAKUT), but current methods are limited by the lack of automated processes that accurately classify diseased and normal kidneys.ObjectiveThe objective of the study was to evaluate the diagnostic performance of deep transfer learning techniques to classify kidneys of normal children and those with CAKUT.Study designA transfer learning method was developed to extract features of kidneys from ultrasound images obtained during routine clinical care of 50 children with CAKUT and 50 controls. To classify diseased and normal kidneys, support vector machine classifiers were built on the extracted features using (1) transfer learning imaging features from a pretrained deep learning model, (2) conventional imaging features, and (3) their combination. These classifiers were compared, and their diagnosis performance was measured using area under the receiver operating characteristic curve (AUC), accuracy, specificity, and sensitivity.ResultsThe AUC for classifiers built on the combination features were 0.92, 0.88, and 0.92 for discriminating the left, right, and bilateral abnormal kidney scans from controls with classification rates of 84%, 81%, and 87%; specificity of 84%, 74%, and 88%; and sensitivity of 85%, 88%, and 86%, respectively. These classifiers performed better than classifiers built on either the transfer learning features or the conventional features alone (p < 0.001).DiscussionThe present study validated transfer learning techniques for imaging feature extraction of ultrasound images to build classifiers for distinguishing children with CAKUT from controls. The experiments have demonstrated that the classifiers built on the transfer learning features and conventional image features could distinguish abnormal kidney images from controls with AUCs greater than 0.88, indicating that classification of ultrasound kidney scans has a great potential to aid kidney disease diagnosis. A limitation of the present study is the moderate number of patients that contributed data to the transfer learning approach.ConclusionsThe combination of transfer learning and conventional imaging features yielded the best classification performance for distinguishing children with CAKUT from controls based on ultrasound images of kidneys.

Project description:Importance:The recent advances in the field of machine learning have raised expectations that computer-aided diagnosis will become the standard for the diagnosis of melanoma. Objective:To critically review the current literature and compare the diagnostic accuracy of computer-aided diagnosis with that of human experts. Data Sources:The MEDLINE, arXiv, and PubMed Central databases were searched to identify eligible studies published between January 1, 2002, and December 31, 2018. Study Selection:Studies that reported on the accuracy of automated systems for melanoma were selected. Search terms included melanoma, diagnosis, detection, computer aided, and artificial intelligence. Data Extraction and Synthesis:Evaluation of the risk of bias was performed using the QUADAS-2 tool, and quality assessment was based on predefined criteria. Data were analyzed from February 1 to March 10, 2019. Main Outcomes and Measures:Summary estimates of sensitivity and specificity and summary receiver operating characteristic curves were the primary outcomes. Results:The literature search yielded 1694 potentially eligible studies, of which 132 were included and 70 offered sufficient information for a quantitative analysis. Most studies came from the field of computer science. Prospective clinical studies were rare. Combining the results for automated systems gave a melanoma sensitivity of 0.74 (95% CI, 0.66-0.80) and a specificity of 0.84 (95% CI, 0.79-0.88). Sensitivity was lower in studies that used independent test sets than in those that did not (0.51; 95% CI, 0.34-0.69 vs 0.82; 95% CI, 0.77-0.86; P?<?.001); however, the specificity was similar (0.83; 95% CI, 0.71-0.91 vs 0.85; 95% CI, 0.80-0.88; P?=?.67). In comparison with dermatologists' diagnosis, computer-aided diagnosis showed similar sensitivities and a 10 percentage points lower specificity, but the difference was not statistically significant. Studies were heterogeneous and substantial risk of bias was found in all but 4 of the 70 studies included in the quantitative analysis. Conclusions and Relevance:Although the accuracy of computer-aided diagnosis for melanoma detection is comparable to that of experts, the real-world applicability of these systems is unknown and potentially limited owing to overfitting and the risk of bias of the studies at hand.

Dataset Information

Improving model fairness in image-based computer-aided diagnosis.

Publications

Improving model fairness in image-based computer-aided diagnosis.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets