View Proposal
- Proposer
- Heba Elshimy
- Title
- Automated CT Body Composition Analysis for Sarcopenia Risk Assessment
- Goal
- To design, implement and rigorously evaluate an automated pipeline that segments skeletal muscle and adipose tissue from single-slice abdominal CT at the third lumbar vertebra (L3), and to determine whether a pre-trained deep learning segmenter offers a clinically meaningful improvement over a classical Hounsfield Unit (HU) thresholding baseline for sarcopenia risk classification.
- Description
- The student will build an end-to-end pipeline that takes an abdominal CT volume as input and outputs a sarcopenia risk flag, with every intermediate quantity (tissue masks, cross-sectional areas, skeletal muscle index) exposed for evaluation. The work compares two segmentation strategies across three tissue compartments and validates them against both segmentation ground truth and a downstream clinical decision.
- Resources
- Data: SAROS (The Cancer Imaging Archive): a large, heterogeneous CT body-composition segmentation dataset with muscle and fat annotations, suitable as primary ground truth. TotalSegmentator dataset (Zenodo): CT volumes with anatomical labels including vertebrae and tissue types, useful for L3-localisation validation and as a secondary source.
- Background
- Python, ML, computer vision, pytorch, matplotlib, pandas, numpy
- Url
- External Link
- Difficulty Level
- Challenging
- Ethical Approval
- None
- Number Of Students
- 1
- Supervisor
- Heba Elshimy
- Keywords
- machine learning, deep learning, computer vision, medical imaging, segmentation
- Degrees