Time-Series Analysis of Cancer Progression Using LSTM Network

Cancer progression varies; in its simplest definition, it is the growth of abnormal cells anywhere on or within an organism. Time-series cancer data are high dimensional and non-linear, demanding a more advanced approach than traditional statistical tools. Long Short-Term Memory (LSTM) networks have become popular for time-series analysis as they can accommodate long-term dependencies well and model complex non-linear patterns. Here, we suggest using LSTM networks in cancer progression data that are often time-dependent and collected for each patient. To accurately identify discriminator signals in the time-series data, we present a stacked LSTM architecture using a deep learning framework generating adversarial examples and feature selection. We have shown that our proposed LSTM approach is superior to common statistical techniques in cancer progression prediction and feature selection. The new cancer progression LSTM network developed a time-series analysis, which has the possibility of being more effective and understandable in monitoring cancer progression for designing offer diagnosis along with treatment compared to the min max AI system.