Dr. Answer AI Software for Prostate Cancer: Explainable Variable Importance of Predicting T Stage

Dr. Answer AI software project for prostate cancer was conducted in South Korea. One of the things we need to develop an AI software model is creating a model that can be explained. We calculated the importance of variables of predicting T stage before developing the Dr. Answer AI software model and used it. We collected 7,128 cases clinical data of prostate cancer patients after the radical prostatectomy treatment: 1,723 were from 2008 to 2017. It is from hospital C, 2,751 from hospital S, and 2,654 from hospital A. We used the random forest algorithm to calculated the importance of variables. We choose the SMOTE+ENN to handle imbalance data sets. Accuracy with SMOTE+ENN is 90.7%. The top important variables in the T stage prediction were 1) initial PSA, 2) BMI, 3) max positive core count, 4) Gleason group, and 5) core ratio. We provide top important variables in T stage prediction to develop an AI software is creating a model that can be explained. Our research can provide guidelines in developing AI SW for prostate cancer.