Related Experiment Video
Updated: Feb 24, 2026
11:13
Quantification of Mouse Heart Left Ventricular Function, Myocardial Strain, and Hemodynamic Forces by Cardiovascular Magnetic Resonance Imaging
Published on: May 24, 2021
7.3K
Automatic detection of hemorrhagic pericardial effusion on PMCT using deep learning - a feasibility study
Lars C Ebert1, Jakob Heimer2, Wolf Schweitzer2
1Institute of Forensic Medicine, University of Zurich, Winterthurerstrasse 190/52, 8057, Zurich, Switzerland. Lars.ebert@virtopsy.com.
Forensic Science, Medicine, and Pathology
|August 19, 2017
Summary
Deep learning can automatically detect hemopericardium in post mortem computed tomography (PMCT) scans. This AI approach shows potential for forensic imaging analysis, aiding in identifying non-natural causes of death.
Area of Science:
- Forensic Radiology
- Medical Imaging Analysis
- Artificial Intelligence in Medicine
Background:
- Post mortem computed tomography (PMCT) serves as a crucial triage tool for identifying potential non-natural deaths, especially with high autopsy caseloads.
- High-resolution imaging in forensic settings generates substantial data, necessitating efficient analysis methods.
- Deep learning (DL) techniques offer a potential solution for automating the analysis of complex radiological images.
Purpose of the Study:
- To evaluate the feasibility of using deep learning methods for automated detection and segmentation of hemopericardium in PMCT images.
- To test the hypothesis that DL algorithms can accurately identify hemopericardium in forensic radiological data.
Main Methods:
- Retrospective analysis of 28 cases with hemopericardium and 24 without, using ViDi Suite 2.0 for deep learning image analysis.
- Training two separate DL networks: one for classification (hemopericardium/not hemopericardium) and one for segmentation (blood content).
- Utilizing a 50/50 train-validation split, repeated 20 times, to assess network performance.
Main Results:
- The best classification network achieved accurate identification of all hemopericardium cases in validation data, with minimal false positives.
- The best segmentation network demonstrated a tendency to underestimate the volume of hemopericardium, a common challenge in such algorithms.
- This study represents the first investigation into the application of deep learning for automated radiological image analysis in forensic medicine.
Conclusions:
- Deep learning models show significant potential for the automated analysis of radiological images in forensic medicine.
- Automated detection and segmentation of findings like hemopericardium using DL could enhance the efficiency of PMCT interpretation.
- Further development of DL algorithms is warranted to improve accuracy and reliability in forensic imaging applications.
