Predicting drug concentrations from forensic autopsy cases using machine learning

Sammanfattning

When determining the cause of death, the Swedish National Board of Forensic Medicine performs mass spectrometry analysis on post-mortem samples to identify the presence of any drugs. This analysis process takes a significant amount of time, and this process may be possible to speed up using deep learning. Deep learning has emerged as a powerful tool for making accurate predictions on complex data in a wide range of fields. In this study, three deep learning and two non-deep learning models were trained and tested to predict drug concentrations from LC-QToF mass spectrometry data and XCMS processed data. The results show that machine learning can predict drug concentrations from mass spectrometry data with varying accuracy. Of the five models tested, the non-deep learning models performed better than the deep learning models. All five models generally had more accurate predictions at lower drug concentration ranges. Due to the importance of reliable results in forensic toxicology, more work is needed before being used as part of determining drug concentrations in post-mortem samples.