Endoscopic Diagnostic Support System for cT1b Colorectal Cancer Using Deep Learning

Area of Science:

• Gastroenterology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate diagnosis of early-stage colon cancer, particularly differentiating between Tis, T1a, and T1b stages, is crucial for effective treatment.
• Endoscopic diagnosis relies heavily on the skill and experience of the endoscopist, leading to potential variability.

Purpose of the Study:

• To evaluate the efficacy of a convolutional neural network (CNN) as a deep learning tool for assisting in the diagnosis of colon cancer stage T1b (cT1b).

Main Methods:

• A retrospective analysis of 190 unenhanced colonoscopy images from 41 patients was performed.
• AlexNet and Caffe deep learning frameworks were utilized for machine learning analysis.
• Data augmentation, oversampling, and 3-fold cross-validation were employed to train and validate the CNN model.

Main Results:

• The CNN model achieved a sensitivity of 67.5%, specificity of 89.0%, and accuracy of 81.2% for cT1b diagnosis.
• The area under the receiver operating characteristic curve (AUC) was calculated to be 0.871, indicating good diagnostic performance.

Conclusions:

• A machine learning-based endoscopic diagnostic support system, utilizing CNNs, enables quantitative diagnosis of colon cancer stages.
• This AI system can provide objective evaluations, minimizing subjectivity and dependence on individual endoscopist expertise.