Using LSTM Networks for Predicting Metastasis in Breast Cancer Patients

Metastasis, which is the dissemination of cancer cells to distant sites in the body, remains one of the leading causes that take most lives among breast cancer patients since it becomes refractory to current therapies once it progresses. An early prediction of metastasis is essential for better patient outcomes and treatment. Our study aims to predict breast cancer patients' metastasis using Extended Immediate Memory (LSTM) systems, which are one type of RNN. The LSTM network has recently become increasingly popular in sequence modeling and performed well on applications like NLP or time series prediction. Extensive evaluation on five real-world breast cancer datasets, which contain clinical and molecular features (e.g., age at diagnosis, tumor size, hormone receptor status, molecular subtype) compared to normal tissue as control samples. The network will learn the relationships between these variables and their effect on which patients are most likely to get metastasis. Using the trained LSTM network, the prediction of metastasis in breast cancer patients is determined. The performance is compared against standard statistical models to study the ability of the LSTM network. The study aims to illustrate the potential application of LSTM networks in metastasis prediction and its advantages compared with common methods. If validated, the LSTM network would be a potential clinical tool to distinguish high-risk breast cancer patients eligible for more aggressive treatment. In the end, this will help in a better prognosis for patients and personalized strategy options of treatment for these breast cancer survivors.