Statistical shape model of a liver for autopsy imaging

Area of Science:

• Medical Imaging
• Anatomical Modeling
• Forensic Science

Background:

• Accurate postmortem liver modeling for autopsy imaging is hindered by organ deformation variability and scarce CT scan data.
• Statistical Shape Models (SSMs) are crucial for analyzing anatomical variations but require robust training datasets.

Purpose of the Study:

• To develop and evaluate an algorithm for constructing a statistical shape model (SSM) of the adult postmortem liver for autopsy imaging.
• To address the challenges posed by postmortem organ deformation and limited data availability.

Main Methods:

• Investigated the relationship between in vivo and postmortem liver SSMs using level set functions and spatially weighted principal components analysis.
• Developed and compared algorithms for transforming in vivo liver shapes to postmortem liver shapes to synthesize training data.
• Constructed and evaluated five postmortem liver SSMs using synthesized and actual postmortem liver labels, comparing them against conventional SSMs.

Main Results:

• SSMs trained solely on in vivo liver data performed poorly in describing postmortem liver shapes.
• Two novel SSMs, trained on synthesized postmortem liver data, demonstrated superior performance and statistical significance over conventional SSMs.
• The proposed methods outperformed traditional SSMs trained on postmortem and/or in vivo liver data.

Conclusions:

• The performance of SSMs for postmortem liver analysis can be significantly improved by incorporating synthesized shape labels derived from in vivo data.
• Combining actual postmortem liver shape labels with synthesized labels offers a viable strategy to enhance SSM accuracy in autopsy imaging.