Artificial intelligence algorithms for cancer prediction through breath analysis.

Our research aims at studying the onset of cancer diseases through the analysis of breath. By determining the abundance of around 250 molecules present in the human breath, obtained through gas chromatography -- mass spectrometry, we have studied the cases of patients affected by prostate and breast cancer. After a preliminary selection process of the most important molecules for the discrimination between healthy controls and affected subjects, we have implemented a Random Forest Machine Learning algorithm. Applying the proper statistical tests, consisting in 10-fold cross validation repeated 100 times, we have quantified the performance of the considered machine learning algorithm. The results indicate very good performances both for the healthy/affected discrimination of breast cancer (accuracy 75%, specificity 80%, sensitivity 68%) and for the three considered levels, classified according to PI-RADS values, in the case of prostate cancer (accuracy 71%).