A symptom based machine learning model for the prediction of prostate cancer:a case study of Uganda Cancer Institute- Mulago.

Abstract

Introduction: Artificial Intelligence (AI) has gained popularity globally as a way of automating previously manual tasks to obtain efficiency at a minimal cost. The health sector is not an exception to this trend. Machine Learning (ML), which is a branch of AI has been employed to make good use of the ever-increasing pool of data in healthcare to develop algorithms or models that can accurately predict the occurrence of future medical events. Objective: The intention of this study was to identify common key symptoms and characteristics and use them to develop and test a model that can be used in the prediction of prostate cancer. This model could be deployed in AI diagnostic tools within the Ugandan health sector and beyond. Methods: Comparison was made between machine learning modelsthat included Naïve Bayes, Decision Tree, K-Nearest Neighbour (KNN), and Logistic regression that were developed using common patient key complaints and characteristics. The performance of these models in prostate cancer prediction was evaluated against gold standard laboratory test results for prostate cancer. The confusion matrix generated from the different models and the Area Under Curve (AUC) was used to make comparisons. Results: All the models achieved a sensitivity, specificity, AUC and Kappa statistic above 80%. Conclusion: The developed models performed sufficiently well and thus can effectively be deployed in screening tools for prostate cancer.