Cancer survival rate prediction using residual neural network on 3D non-spatial data

• Main Author: Chua, Yue Da
• Other Authors: Cai Yiyu
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/138613
• Institution: Nanyang Technological University

Improved cancer prognosis is an important goal of precision health medicine. Triple negative breast cancer (TNBC), being an aggressive form of cancer, requires novel and effective treatment. Deep Learning and its ability to model complex data inputs presents itself as a useful candidate for this application, which we hope could provide deeper insights and provide patients with more reliable results. We apply deep learning to patient’s clinical data and Multiplex Immunohistochemistry Image numerical data to determine a precise survival rate prediction using scalar regression. We accomplish this using state of the art Deep Learning architectures of Wide Residual Network and Residual Network with Identity Mapping with the aim of using Residual Network highways as a means of learning meaningful representation of the data while building a model complex enough to generalize the dataset. Our model achieved a mean absolute error of 25.86 months. Throughout the span of the project, we also built on past progress by evaluating our model in a more robust manner via K-Fold Cross Validation, and also explore further data preprocessing process to ensure a more accurate result.Through our study, we gained a good understanding of dealing with the less researched, but prevalent 3D non-spatial dataset type that is unique to our case, and applying it to 2D CNN, which is predominantly used for 3D spatial datasets like images. We used our experience to developed a rule of thumb framework for dealing with a 3D non-spatial data and leveraging the vast work done in 2D CNN to solve complex real-world problems, and also as a means of solving the aim of providing accurate prediction on the survival rate of TNBC patients for better prognosis in the future.