Smart Diagnosis of Cholangiocarcinoma from Microscopic Images Using a Modified Visual Geometry Group Network with Adaptive Augmentation

Hyperspectral Microscopic Images (HSMI) offer a more comprehensive spectral range, making them particularly useful in medical diagnostics, including early detection of cancerous tissues. Cholangiocarcinoma, a highly lethal bile duct cancer, traditionally requires a histopathological analysis of tissue samples under a microscope. However, this method is prone to subjectivity and error, often leading to delays in diagnosis, contributing to patient mortality. This study focuses on the automated diagnosis of cholangiocarcinoma using deep learning techniques on microscopic hyperspectral data. Leveraging the spectral richness of HSMI, we developed a framework utilizing a modified Visual Geometry Group (VGG) architecture to process this hyperspectral data, aiming to detect cholangiocarcinoma at its early stages. With the growing role of artificial intelligence in pathology, this approach minimizes human error and enhances diagnostic accuracy. The dataset used in this study includes 880 cholangiocarcinoma tissue samples from 174 individuals, comprising 689 partial cancer regions, 49 full cancer regions, and 142 healthy scenes, all meticulously labeled by expert pathologists. Our ensemble learning technique integrates image preprocessing, spectral feature extraction, and classification, significantly improving diagnostic accuracy. The proposed system demonstrates a substantial improvement in the early detection of cholangiocarcinoma and offers a valuable tool for smart microscopy-based diagnosis, potentially facilitating the diagnostic burden in clinical settings.