DL-based segmentations on CT and MRI show limited agreement and bias compared to manual segmentations, affecting radiomics model reliability.
Key Details
- 1Study evaluated manual vs. DL-generated segmentation in abdominal CT (139 cases) and MR (64 cases) datasets in pediatric and young adult populations.
- 2Multiple open-source DL segmentation tools were compared for each imaging modality and dataset.
- 3A significant number of radiomic features demonstrated poor or fair agreement and bias when derived from DL-generated vs. manual segmentations.
- 4Model accuracy and generalizability often worsened when using features from DL segmentations, especially for MRSegmentator and PaNSegNet.
- 5Authors advise full transparency regarding segmentation methods in radiomics research, as methodological choices can confound results.
Why It Matters
Segmentation method choice can significantly influence radiomic analyses in both CT and MRI, impacting reproducibility and comparability of AI models in research and clinical practice. Clear reporting and awareness of these methodological effects are essential for advancing imaging AI.
Source
AuntMinnie
Related News
•Radiology Business
A Practical Guide to Evaluating Radiology AI Solutions
Experts propose a three-step approach for radiologists to assess and select AI imaging solutions.
•AuntMinnie
AI Quantitative MRI and Amyloid PET Used for ARIA Monitoring in Alzheimer's
AI-driven MRI analysis complements Amyloid PET in detecting ARIA during Alzheimer's treatment.
•Radiology Business
New Framework Adapts Adult Radiology AI Models for Pediatric Use
Researchers created a privacy-preserving method to adapt adult-trained radiology AI models for use in pediatric populations.