Cancer Metastasis Prediction for Effective Blocking via Backpropagation

摘要

Due to the exponential growth of cancer cell multiplication, early detection to stop cancer spreading to vital organs is the key to effective cancer treatment. Visualizing existing cancer metastasis data to predict the cancer spreading paths toward vital organs can provide more clues and broader confidence for the physicians to make better treatment decisions. In additional to treating or removing cancer tumors, the patients can maximize their chance to prolong lives by also treating to block the most likely cancer spreading paths to vital organs. In our backpropagation (back-prop) artificial neural network (ANN) model experiments, we aim to (1) summarize and visualize existing cancer metastasis data for clear, meaningful references by doctors, (2) predict the probability of likely spreading paths, and (3) provide favorable results as evidence to urge our health care sectors to gather, preserve and made available more metastasis data. Eventually, we hope that doctors can have thousands to millions of cancer metastasis cases under their finger tips with categorized variations of cancer spreading patterns. The possible cancer metastasis factors are cancer types, genes, chemical signals, hormones, age, gender, culture, diet, etc. Our preliminary back-prop results successfully predict the British head and neck cancer spreading sites with noted probabilities. The result can be easily verified because that the cancer metastasis pattern in the British head and neck cancer seems to be a more predictable type of cancer spreading, closely following the previous statistic analysis as well as the physical proximity. With this progress, we are looking to obtain and extend our experiment on other cancer metastasis data that tend to show cancer spreading in a less obvious manner.