Use Test of Automated Machine Learning in Cancer Diagnostics

Our aim is to investigate the added value of automated machine learning (AutoML) for potential future applications in cancer diagnostics. Using two important diagnostic questions, non-invasive determination IDH mutation status and ATRX status, we analyze whether it possible use AutoML develop models that are comparable performance conventional (ML) developed by experts. For this purpose, using different feature preselection methods compare results with previously ML models. The cohort used our study comprises T2-weighted MRI images 124 patients histologically confirmed gliomas. AutoML, were able sophisticated a very short time only few lines computer code. In predicting obtained mean AUC 0.7400 AUPRC 0.8582. was predicted similar discriminatory power, 0.7810 0.8511. both cases, even achieve power slightly above respective conventionally computing time, thus making such accessible non-experts near future.