RESUMO
Background: Prostate cancer (PCa) is among the most common cancers in men and its diagnosis requires the histopathological evaluation of biopsies by human experts. While several recent artificial intelligence-based (AI) approaches have reached human expert-level PCa grading, they often display significantly reduced performance on external datasets. This reduced performance can be caused by variations in sample preparation, for instance the staining protocol, section thickness, or scanner used. Another limiting factor of contemporary AI-based PCa grading is the prediction of ISUP grades, which leads to the perpetuation of human annotation errors. Methods: We developed the prostate cancer aggressiveness index (PCAI), an AI-based PCa detection and grading framework that is trained on objective patient outcome, rather than subjective ISUP grades. We designed PCAI as a clinical application, containing algorithmic modules that offer robustness to data variation, medical interpretability, and a measure of prediction confidence. To train and evaluate PCAI, we generated a multicentric, retrospective, observational trial consisting of six cohorts with 25,591 patients, 83,864 images, and 5 years of median follow-up from 5 different centers and 3 countries. This includes a high-variance dataset of 8,157 patients and 28,236 images with variations in sample thickness, staining protocol, and scanner, allowing for the systematic evaluation and optimization of model robustness to data variation. The performance of PCAI was assessed on three external test cohorts from two countries, comprising 2,255 patients and 9,437 images. Findings: Using our high-variance datasets, we show how differences in sample processing, particularly slide thickness and staining time, significantly reduce the performance of AI-based PCa grading by up to 6.2 percentage points in the concordance index (C-index). We show how a select set of algorithmic improvements, including domain adversarial training, conferred robustness to data variation, interpretability, and a measure of credibility to PCAI. These changes lead to significant prediction improvement across two biopsy cohorts and one TMA cohort, systematically exceeding expert ISUP grading in C-index and AUROC by up to 22 percentage points. Interpretation: Data variation poses serious risks for AI-based histopathological PCa grading, even when models are trained on large datasets. Algorithmic improvements for model robustness, interpretability, credibility, and training on high-variance data as well as outcome-based severity prediction gives rise to robust models with above ISUP-level PCa grading performance.
RESUMO
Purpose: To develop, train, and validate a multiview deep convolutional neural network (DeePSC) for the automated diagnosis of primary sclerosing cholangitis (PSC) on two-dimensional MR cholangiopancreatography (MRCP) images. Materials and Methods: This retrospective study included two-dimensional MRCP datasets of 342 patients (45 years ± 14 [SD]; 207 male patients) with confirmed diagnosis of PSC and 264 controls (51 years ± 16; 150 male patients). MRCP images were separated into 3-T (n = 361) and 1.5-T (n = 398) datasets, of which 39 samples each were randomly chosen as unseen test sets. Additionally, 37 MRCP images obtained with a 3-T MRI scanner from a different manufacturer were included for external testing. A multiview convolutional neural network was developed, specialized in simultaneously processing the seven images taken at different rotational angles per MRCP examination. The final model, DeePSC, derived its classification per patient from the instance expressing the highest confidence in an ensemble of 20 individually trained multiview convolutional neural networks. Predictive performance on both test sets was compared with that of four licensed radiologists using the Welch t test. Results: DeePSC achieved an accuracy of 80.5% ± 1.3 (sensitivity, 80.0% ± 1.9; specificity, 81.1% ± 2.7) on the 3-T and 82.6% ± 3.0 (sensitivity, 83.6% ± 1.8; specificity, 80.0% ± 8.9) on the 1.5-T test set and scored even higher on the external test set (accuracy, 92.4% ± 1.1; sensitivity, 100.0% ± 0.0; specificity, 83.5% ± 2.4). DeePSC outperformed radiologists in average prediction accuracy by 5.5 (P = .34, 3 T) and 10.1 (P = .13, 1.5 T) percentage points. Conclusion: Automated classification of PSC-compatible findings based on two-dimensional MRCP was achievable and demonstrated high accuracy on internal and external test sets.Keywords: Neural Networks, Deep Learning, Liver Disease, MRI, Primary Sclerosing Cholangitis, MR Cholangiopancreatography Supplemental material is available for this article. © RSNA, 2023.