Hierarchical Text Classification of Autopsy Reports to Determine MoD and CoD Through Term-Based and Concepts-Based Features

Abstract

Nowadays, text classification has been extensively employed in medical domain to classify free text clinical reports. In this study, text classification techniques have been used to determine cause of death from free text forensic autopsy reports using proposed term-based and SNOMED CT concept-based features. In this study, detailed term-based features and concept-based features were extracted from a set of 1500 forensic autopsy reports belonging to four manners of death and 16 different causes of death. These features were used to train text classifier. The classifier was deployed in cascade architecture: the first level will predict the manner of death and the second level will predict the CoD using proposed term-based and SNOMED CT concept-based features. Moreover, to show the significance of our proposed approach, we compared the results of our proposed approach with four state-of-the-art feature extraction approaches. Finally, we also presented the comparison of one-level classification versus two-level classification. The experimental results showed that our proposed approach showed 8% improvement in accuracy as compared to other four baselines. Moreover, two-level classification showed improved accuracy in determining CoD compared to one-level classification.