Development of artificial intelligence based web interface support system with R-Shiny application from open source clinical cancer data

Accurate prognosis of cancer patients is important for optimizing treatment strategies, and designing clinical trials. This study was conducted to estimate the survival of the patients diagnosed with colon cancer using machine learning techniques based on classification. The study included clinical and genomic data of 454 colon cancer patients from the NCI "TCGA-COAD" project on the open access GDC portal. Clinical and genomic parameters that were found to be significantly associated with survival were used as risk groups. Analyzes were carried out with codes written in R language. Estimates were made using risk groups as class labels. Selected clinical and genomic parameters; was evaluated with four different algorithms. Model evaluation was performed by creating test data. The prediction model was transformed into AI-based CDSS (Clinical Decision Support System) using the web interface via Shiny. According to the results of the study; It was seen that the machine learning algorithm that predicted the most accurate survival was the Random Forest. In addition, it was seen that the use of all clinical and genomic data together increased the algorithm performance (76.2%). AI-based CDSS models that are focus on survival outcome in the field of oncology are important in terms of providing foresight for physicians and patients about whether it is worth applying the chemotherapy treatment which might seriously affect the patients. Developing and improving such systems will make the management of the disease easier and help make more accurate decisions for the patient.