Radiomics-Based Quality Control System for Automatic Cardiac Segmentation: A Feasibility Study.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging Analysis

Background:

• Automatic cardiac segmentation methods have advanced rapidly, yet automatic quality control (QC) remains underdeveloped.
• Ensuring the accuracy of cardiac segmentation is crucial for reliable cardiovascular analysis.

Purpose of the Study:

• To develop an automated pipeline integrating deep learning (DL) for cardiac segmentation and radiomics for quality control (QC).
• To address the gap in automatic QC for DL-based cardiac segmentation methods.

Main Methods:

• A DL-based approach was used for heart localization and segmentation of cardiac substructures (right ventricle cavity, myocardium, left ventricle cavity).
• A radiomics dataset was created for feature extraction, selection, and development of a QC model.
• The pipeline was evaluated using internal and external testing datasets.

Main Results:

• The segmentation model achieved high Dice Similarity Coefficient (DSC) values (e.g., 0.954 for whole heart, 0.940 for LVC 2D, 0.962 for LVC 3D).
• Radiomics-based QC models accurately predicted segmentation quality with low mean absolute errors (e.g., 0.021 for LVC 2D, 0.011 for LVC 3D).
• High negative detection rates (>0.85) were observed for 2D segmentation QC, with similar performance on external datasets.

Conclusions:

• An automated pipeline for cardiac substructure segmentation and QC at both 2D and 3D levels was successfully developed.
• Radiomics demonstrates strong potential as an automated method for quality control in cardiac segmentation.