Test of a new method in the distinction between falls and blows on a postmortem CT-scans sample.

Abstract

The discrimination between falls and blows is an important task in forensic anthropology and pathology. This research aimed to test a discrimination method between falls and blows. This method was created from the quotation of 549 types of fractures for 57 bones and 12 anatomical regions. Different models were tested according to the sensibility of random forest parameters and their effects on model accuracies. The best model was based on binary coding of 12 anatomical regions or 28 bones with or without baseline (age and sex). We tested this new method in the distinction between falls and blows on post-mortem computerized tomography scans (PMCT). The sample was composed of 47 subjects with 36 falls and 11 blows, whose aetiologia was based on forensic reports. Of the 47 bodies, 35 were complete, and 12 presented missing bones; 39 were estimated to be falls and 8 to be blows. Of the 12 individuals with missing bones, 11 had a reasonable estimation of the etiology of fractures, i.e., 91.7%. Methods showed excellent etiology estimation for fall cases (97.2%) but misclassified 36.4% of blow cases. Our method misclassified 5 subjects (10.6%), more precisely, 4 blows and 1 fall. Overall, the reliability of the estimation of the etiology is substantial, with a Cohens k-values of 0.67. The method could be used in distinguishing between blows and falls and is also suitable for fragmented or missing bones. To ensure an easy and fast use of this method, we have developed a freely available online automated tool (http://fracture.cloud).