A Deep Learning Aided Drowning Diagnosis for Forensic Investigations using Post-Mortem Lung CT Images

Area of Science:

• Forensic Medicine
• Radiology
• Artificial Intelligence

Background:

• Computer-aided diagnosis (CAD) systems, particularly those using deep learning, have demonstrated significant capabilities in medical image analysis.
• However, the application of CAD systems to post-mortem imaging, especially for challenging diagnoses like drowning, remains unexplored.
• Forensic medicine faces difficulties in diagnosing drowning due to the non-specific nature of post-mortem imaging findings.

Purpose of the Study:

• To develop and evaluate a deep convolution neural network (DCNN) based CAD system for classifying post-mortem lung CT images.
• To determine the effectiveness of a DCNN in distinguishing between drowning and non-drowning cases using post-mortem CT data.
• To assess the potential of AI in improving the accuracy of drowning diagnosis in forensic investigations.

Main Methods:

• A DCNN was developed for classifying post-mortem lung CT images into drowning and non-drowning categories.
• The DCNN model was trained using transfer learning techniques.
• Performance was evaluated using 10-fold cross-validation on a dataset of 140 drowning and 140 non-drowning cases.

Main Results:

• The DCNN-based CAD system achieved an average area under the receiver operating characteristic curve (AUC-ROC) of 0.88.
• This performance indicates a high level of accuracy in classifying post-mortem lung CT scans for drowning.
• The study demonstrates the feasibility of using deep learning for this specific forensic diagnostic task.

Conclusions:

• The proposed DCNN-based CAD system shows significant potential for aiding in the diagnosis of drowning from post-mortem CT imaging.
• This AI-driven approach can potentially enhance the accuracy and efficiency of forensic investigations.
• Further development and validation of CAD systems in post-mortem imaging could revolutionize forensic pathology.