On simulating subjective evaluation using combined objective metrics for validation of 3D tumor segmentation

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Radiology

Background:

• Accurate 3D tumor segmentation in CT images is crucial for diagnosis and treatment planning.
• Current objective segmentation evaluation metrics may not fully capture radiologist's subjective assessment.
• Developing reliable automated evaluation methods is essential for advancing segmentation algorithms.

Purpose of the Study:

• To develop a novel segmentation evaluation method that simulates radiologist's subjective assessment of 3D tumor segmentation in CT images.
• To introduce a new composite metric that combines commonly used objective metrics.
• To validate the proposed method's performance against radiologist ratings and individual objective metrics.

Main Methods:

• A new composite metric was defined as a linear combination of established objective segmentation metrics.
• Weighting parameters for the composite metric were optimized by maximizing rank correlation with radiologist subjective ratings.
• The method was evaluated on 93 lesions from CT images, comparing the composite metric's performance against individual objective metrics and radiologist assessments.

Main Results:

• The proposed composite metric demonstrated superior performance in segmentation evaluation compared to individual objective metrics.
• Segmentation ratings derived from the composite metric showed strong agreement with subjective evaluations by radiologists.
• The method effectively simulated radiologist's subjective assessment of tumor segmentation quality.

Conclusions:

• The developed composite metric offers a more reliable and clinically relevant approach to evaluating 3D tumor segmentation in CT images.
• This method has the potential to accelerate the development of advanced tumor segmentation algorithms.
• The approach can support large-scale segmentation evaluation studies and improve diagnostic accuracy.