M.Sc Thesis


M.Sc StudentRozen Rakefet
SubjectMachine Learning Model based on Innovative,
Mechanobiology Assay for Patient-Specific
Prediction of Metastatic Risk
DepartmentDepartment of Biomedical Engineering
Supervisor ASSOCIATE PROF. Yael Yaniv
Full Thesis textFull thesis text - English Version


Abstract

Cancer mortality is mostly related to metastasis. Metastasis is currently prognosed via histopathology, disease statistics, or genetics; those are not infallible, are not rapidly available, and require knowledge of specific markers. The Weihs lab in the Technion have developed a rapid (~2hr) mechanobiology-based approach to provide early prognosis of the clinical risk for tumor metastasis by evaluating the invasiveness of sampled cells. Specifically, invasive subsets of cell samples that are seeded on impenetrable, physiological-stiffness polyacrylamide gels will forcefully push and indent the gels, while non-invasive cells (e.g. benign) do not. The number of indenting cells and their attained depths provides the mechanical invasiveness of the sample, which they have shown agrees with the in vitro metastatic potential of cell lines and in vivo metastatic risk of various types of tumors. Utilizing the experimental database, we have compared the capacity of several machine learning models to predict the metastatic risk. The models underwent supervised training on individual experiments using classification from literature and commercial-source information for established cell lines and clinical histopathology reports for tumor samples. We evaluated two classification approaches by dividing the data to 2 or 3 classes. With 2-class models separating non-invasive (i.e. normal, benign and non-metastatic cancer) and invasive (metastatic cancer) data, we obtained sensitivity and specificity of 0.97, exceeding other published classification attempts in this field. Introducing a novel metastatic prediction approach, we a 3-class model (i.e. no, low or high metastatic risk) that provided sensitivity of 1, 0.71, 0.8 and specificity of 1, 0.88, 0.85 to each class respectively. Combining our rapid, mechanical invasiveness assay with machine learning classification can provide accurate and early prognosis of metastatic risk, to support choice of treatments and improve disease management.