PR-OCT Machine Learning For Colon Cancer Detection

Researchers are developing a deep learning-based pattern recognition, i.e., PR-OCT system, that will automate image processing and provide an accurate, computer-aided diagnosis of colorectal cancer potentially in real-time. The method uses deep learning and OCT machine learning to find patterns in healthy and abnormal tissue sample images.

Using OCT as a study tool, the researchers started imaging patient-removed samples of colorectal tissue in 2017. According to researchers, healthy colorectal tissue was found to have a structure resembling teeth. The precancerous and cancerous cells, however, hardly ever exhibited this pattern. The healthy mucosa microstructures of the colorectal tissue had mildly attenuated, which led to the development of the tooth pattern.

RetinaNet, a neural network model of the brain, was trained to recognize and understand the structural patterns in OCT images of the human colon. About 26,000 OCT images from 20 tumor areas, 16 benign areas, and six other abnormal areas in patient tissue samples were used to train and evaluate the network.

The trained network efficiently recognized the patterns distinguishing healthy and malignant colorectal tissue. The experimental diagnoses suggested by the PR-OCT method were contrasted with assessments of the tissue samples made using conventional histology. In this pilot study, compared to pathology reports, the PR-OCT could detect tumors with 100% accuracy. A 100% sensitivity and 99.7% specificity were attained.

Read more

Related Content: Imaging Method Enables Actin Research