Enhancing Post-Surgical Wound Care in Anorectal Diseases: A Comparative Study of Advanced Convolutional Neural Network (CNN) Architectures for Image Classification and Analysis

Abstract

AIM: Anorectal diseases, often requiring surgical intervention and careful post-operative wound management, pose substantial challenges in healthcare. This study presents a novel application of artificial intelligence, specifically machine learning, aimed at improving the classification and analysis of post-surgical wound images. By doing so, it seeks to enhance patient outcomes through personalized and optimized wound care strategies. 

METHODS: This research utilizes convolutional neural networks (CNNs) and employs three advanced architectures—MobileNet, ResNet50, and Inception-v4—to detect and classify key characteristics of post-surgical wounds, including size, location, severity, and tissue type involved. Additionally, the study integrates Gradient-weighted Class Activation Mapping (Grad-CAM) technology to provide interpretative insights into the decision-making processes of these algorithms, offering a deeper understanding of model predictions. 

RESULTS: The effectiveness of the employed CNN architectures was assessed based on accuracy, precision, and recall metrics. The findings demonstrate that Inception-v4, in particular, exhibits superior performance across all evaluated metrics, underscoring its potential in clinical applications. Grad-CAM visualizations further clarified the rationale behind the model's decisions, enhancing the interpretability of the results. 

CONCLUSIONS: The integration of machine learning technologies in the classification and analysis of wound images represents a significant advancement in medical image analysis and AI-driven healthcare solutions. This research not only enhances the technical capabilities of AI applications in healthcare but also improves the precision of post-operative care in anorectal surgery, ultimately contributing to better treatment outcomes.