Codebase for the final project of CISC 881 - Medical Informatics. Developed a machine learning pipeline to classify tissue types using optical spectrum. Trained and evaluated ML and DL models. Achieved an average F1 score of 95% and an accuracy of 95.5%.
Project abstract: Over 30% of breast-conserving surgery (BCS) patients must undergo revision surgery due to cancerous tissue remaining in the cavity. We hypothesize that optical tissue characterization and classification can improve the success rate of BCS by giving the surgeon more information to work with intraoperatively. In this study, we assess the feasibility of diffuse reflection broadband spectroscopy to differentiate between Beef, Chicken, Pork, and Turkey samples. We recorded the optical reflectivity of various tissue samples in a range of 200 to 1000 nm. We then preprocessed this data and used an SVM, a KNN, and a 1D CNN to classify the resulting spectra. The KNN achieved an average F1 score of 95% and an accuracy of 95.5%. This outperformed the CNN and SVM which achieved 89% and 84% respectively. Through broadband spectroscopy, we demonstrate the feasibility of optical characterization for tissue classification. Applying this to breast conserving surgery can help inform the surgeon about the tissue composition of the resection cavity after initial tumour resection.