Deep-Learning for Classification of Colorectal Polyps on Whole-Slide Images (1703.01550v2)

Abstract: Histopathological characterization of colorectal polyps is an important principle for determining the risk of colorectal cancer and future rates of surveillance for patients. This characterization is time-intensive, requires years of specialized training, and suffers from significant inter-observer and intra-observer variability. In this work, we built an automatic image-understanding method that can accurately classify different types of colorectal polyps in whole-slide histology images to help pathologists with histopathological characterization and diagnosis of colorectal polyps. The proposed image-understanding method is based on deep-learning techniques, which rely on numerous levels of abstraction for data representation and have shown state-of-the-art results for various image analysis tasks. Our image-understanding method covers all five polyp types (hyperplastic polyp, sessile serrated polyp, traditional serrated adenoma, tubular adenoma, and tubulovillous/villous adenoma) that are included in the US multi-society task force guidelines for colorectal cancer risk assessment and surveillance, and encompasses the most common occurrences of colorectal polyps. Our evaluation on 239 independent test samples shows our proposed method can identify the types of colorectal polyps in whole-slide images with a high efficacy (accuracy: 93.0%, precision: 89.7%, recall: 88.3%, F1 score: 88.8%). The presented method in this paper can reduce the cognitive burden on pathologists and improve their accuracy and efficiency in histopathological characterization of colorectal polyps, and in subsequent risk assessment and follow-up recommendations.

• The paper introduces a ResNet-based model that achieves 93% accuracy in classifying five colorectal polyp types.
• It leverages a robust dataset of 1,723 expert-annotated whole-slide images to ensure reliable performance metrics.
• The approach reduces diagnostic workload and offers scalable potential for enhancing colorectal cancer screening.

An Evaluation of Deep-Learning for Automatic Classification of Colorectal Polyps in Whole-Slide Images

The paper entitled "Deep-Learning for Classification of Colorectal Polyps on Whole-Slide Images" presents a novel methodology for streamlining the histopathological characterization of colorectal polyps using deep learning techniques. By implementing this approach, the authors aim to improve both the diagnostic accuracy and efficiency of pathologists' assessments of colorectal polyps, which are key in determining the risk of colorectal cancer.

Methodological Framework

The paper utilizes a convolutional neural network (CNN) architecture based on ResNet, a variant of deep neural networks known for overcoming the vanishing gradient problem and enabling deeper network configurations. This architecture is implemented to automatically classify five types of colorectal polyps—hyperplastic, sessile serrated, traditional serrated adenoma, tubular adenoma, and tubulovillous/villous adenoma—from whole-slide histological images.

A robust dataset consisting of 1,723 H&E-stained whole-slide images from Dartmouth-Hitchcock Medical Center was employed for the model's development. The images were annotated by pathology experts to create a solid foundation for training and validation. The proposed system analyzes these images by breaking them into smaller segments, ensuring efficient handling by the CNN while retaining key information for accurate classification.

Numerical Results

The ResNet-based model exhibited notable performance on independent test data comprising 239 whole-slide images, achieving an accuracy of 93.0%, with precision, recall, and F1 score values at 89.7%, 88.3%, and 88.8% respectively. These results underscore the model's efficacy in accurately identifying and classifying polyps, particularly distinguishing between polyps with similar morphological features.

Discussion and Implications

The integration of advanced deep-learning techniques in histopathological classification offers significant benefits. It reduces the cognitive load on pathologists, improves diagnostic consistency, and offers a scalable solution accommodating large datasets typical in clinical settings. Furthermore, this system holds promise for expanding coverage and enhancing the accuracy of colorectal cancer screenings.

Although promising, the model necessitates further external validation to confirm its generalizability across diverse populations beyond the Dartmouth dataset, as noted by the authors' plans for future collaboration with the New Hampshire Colonoscopy Registry.

The implications of this research stretch beyond colorectal cancer screenings. The methodological framework and deep-learning architecture can potentially be applied to other histopathological evaluation challenges in different cancer types, promising advancements in precision medicine at large.

Conclusion

The research delineated in this paper presents a compelling case for the integration of deep learning systems in pathology, emphasizing its unprecedented potential to advance colorectal cancer screening initiatives. By facilitating a more efficient, reliable assessment of histological slides, this technique holds the potential to significantly impact public health outcomes through early and precise intervention. As such, it offers a substantive contribution to the ongoing development of AI applications within biomedical domains.