Histology-based Prediction of Therapy Response to Neoadjuvant Chemotherapy for Esophageal and Esophagogastric Junction Adenocarcinomas Using Deep Learning
Abstract
Background: Quantifying treatment response to gastroesophageal junction (GEJ) adenocarcinomas is crucial to provide optimal therapeutic strategy. Routinely taken tissue samples provide an opportunity to enhance existing PET/CT-based therapy response evaluation. Our objective was to investigate if deep learning algorithms are capable to predict the therapy response of GEJ patients to neoadjuvant chemotherapy based on histological tissue samples. Methods: This diagnostic study recruited 67 patients with GEJ I-III from the multicentric non-randomized MEMORI trial including 3 German university hospitals TUM (Munich), LMU (Munich), and UME (Essen). All patients underwent baseline PET/CT scans and esophageal biopsy before and 14-21 days after treatment initiation. Treatment response was defined as a ≥ 35% decrease in SUVmax from baseline. Several deep learning algorithms were developed to predict PET/CT-based responders and non-responders to neoadjuvant chemotherapy using digitized histopathological whole slide images. Results: The resulting models were trained on TUM (n=25 pre-therapy, n=47 on-therapy) patients and evaluated on our internal validation cohort from LMU and UME (n=17 pre-therapy, n=15 on-therapy). Compared with multiple architectures, the best pre-therapy network achieves an area under the precision-recall curve (AUPRC) of 0.81 (95% confidence interval (CI), 0.61-1.00), area under the precision-recall curve (AUPRC) of 0.82 (95% CI, 0.61-1.00), balanced accuracy of 0.78 (95% CI, 0.60-0.94), and a Matthews correlation coefficient (MCC) of 0.55 (95% CI, 0.18-0.88). The best on-therapy network achieves an AUROC of 0.84 (95% CI, 0.64-1.00), AUPRC of 0.82 (95% CI, 0.56-1.00), balanced accuracy of 0.80 (95% CI, 0.63-1.00), and MCC of 0.71 (95% CI, 0.38-1.00), solving a task beyond the pathologists’ capabilities. Conclusions: The findings suggest that the networks can predict treatment response using WSI with high accuracy even pre-therapy, suggesting morphological tissue biomarkers. Subject to further validation, this could lead to earlier therapy intensification compared to current PET/CT diagnostic system for non-responder.
Fabian Hörst
PhD Student
Jan Egger
Team Lead AI-guided Therapies
