Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Decoding Future Risk: Deep Learning Analysis of Tubular Adenoma Whole-Slide Images

Ahmed Rahu, Brian Shula, Brandon Combs, Aqsa Sultana, Surendra P. Singh, Vijayan K. Asari, Derrick Forchetti

TL;DR

This study addresses the need for better risk stratification of progression from low-grade tubular adenomas to colorectal cancer by applying a CNN to ROI-annotated whole-slide images. Using an EfficientNetV2S-based classifier trained on 1024×1024 tiles extracted from 40× scanned WSIs, the approach achieves high tile-level discrimination (AUROC) and perfect slide-level accuracy on held-out samples, with Grad-CAM heatmaps highlighting morphologic cues such as nuclear crowding and architectural irregularities in progressors. The results suggest that subvisual histologic features within low-grade dysplasia may signal future cancer risk, offering a path toward personalized post-polypectomy surveillance. However, generalizability remains to be established through multi-institutional validation and cross-scanner robustness studies.

Abstract

Colorectal cancer (CRC) remains a significant cause of cancer-related mortality, despite the widespread implementation of prophylactic initiatives aimed at detecting and removing precancerous polyps. Although screening effectively reduces incidence, a notable portion of patients initially diagnosed with low-grade adenomatous polyps will still develop CRC later in life, even without the presence of known high-risk syndromes. Identifying which low-risk patients are at higher risk of progression is a critical unmet need for tailored surveillance and preventative therapeutic strategies. Traditional histological assessment of adenomas, while fundamental, may not fully capture subtle architectural or cytological features indicative of malignant potential. Advancements in digital pathology and machine learning provide an opportunity to analyze whole-slide images (WSIs) comprehensively and objectively. This study investigates whether machine learning algorithms, specifically convolutional neural networks (CNNs), can detect subtle histological features in WSIs of low-grade tubular adenomas that are predictive of a patient's long-term risk of developing colorectal cancer.

Decoding Future Risk: Deep Learning Analysis of Tubular Adenoma Whole-Slide Images

TL;DR

This study addresses the need for better risk stratification of progression from low-grade tubular adenomas to colorectal cancer by applying a CNN to ROI-annotated whole-slide images. Using an EfficientNetV2S-based classifier trained on 1024×1024 tiles extracted from 40× scanned WSIs, the approach achieves high tile-level discrimination (AUROC) and perfect slide-level accuracy on held-out samples, with Grad-CAM heatmaps highlighting morphologic cues such as nuclear crowding and architectural irregularities in progressors. The results suggest that subvisual histologic features within low-grade dysplasia may signal future cancer risk, offering a path toward personalized post-polypectomy surveillance. However, generalizability remains to be established through multi-institutional validation and cross-scanner robustness studies.

Abstract

Colorectal cancer (CRC) remains a significant cause of cancer-related mortality, despite the widespread implementation of prophylactic initiatives aimed at detecting and removing precancerous polyps. Although screening effectively reduces incidence, a notable portion of patients initially diagnosed with low-grade adenomatous polyps will still develop CRC later in life, even without the presence of known high-risk syndromes. Identifying which low-risk patients are at higher risk of progression is a critical unmet need for tailored surveillance and preventative therapeutic strategies. Traditional histological assessment of adenomas, while fundamental, may not fully capture subtle architectural or cytological features indicative of malignant potential. Advancements in digital pathology and machine learning provide an opportunity to analyze whole-slide images (WSIs) comprehensively and objectively. This study investigates whether machine learning algorithms, specifically convolutional neural networks (CNNs), can detect subtle histological features in WSIs of low-grade tubular adenomas that are predictive of a patient's long-term risk of developing colorectal cancer.
Paper Structure (13 sections, 5 figures, 3 tables)

This paper contains 13 sections, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Histologic Review and Slide Preparation
  4. Image Preprocessing and Annotation
  5. Feature Extraction
  6. Experimental Setup and Results
  7. Model Training and Validation
  8. Augmentation Techniques
  9. Model Training
  10. Statistical Analysis
  11. Experimental Results
  12. Discussion
  13. Conclusion

Figures (5)

  • Figure 1: Sample images of the dataset used: non-progressor group (left) and progressor group (right).
  • Figure 2: Sample Albumentation Tile Augmentations: Representative examples of augmented tiles generated through the Albumentations pipeline. They demonstrate variations in color, contrast, brightness, and sharpness used to improve model robustness and simulate histologic variability across laboratories and slide scanners.
  • Figure 3: AUROC for the Test Tile Set: Tile-level receiver operating characteristic (ROC) curve generated from the held-out test set ($n=40,514$ tiles), demonstrating excellent distinction between progressor and non-progressor tiles.
  • Figure 4: Results for one progressor WSI: (a.) Original WSI (b.) Tile prediction heatmap (c.) Individual tiles (Predicted: Progressor, Target: Progressor) (d.) Histogram of tile probabilities (Target: 1, Predicted: 1)
  • Figure 5: Original Tiles and Grad-CAM Overlays for non-progressor and progressor Groups. Red indicates the highest relevance for both groups. (a) Non-progressor group tile-level Grad-CAM heatmap. (b) Progressor group tile-level Grad-CAM heatmap.