Benchmarking Wilms' tumor in multisequence MRI data: why does current clinical practice fail? Which popular...

Area of Science:

• Pediatric Oncology
• Medical Imaging
• Computational Pathology

Background:

• Wilms' tumor is a common childhood malignancy requiring accurate segmentation for treatment.
• Existing datasets are insufficient for robustly evaluating segmentation methods.
• Current clinical practices for tumor volume assessment may lack precision.

Purpose of the Study:

• To establish the first heterogeneous benchmark dataset for Wilms' tumor segmentation.
• To analyze interrater variability in human expert annotations.
• To evaluate and compare the performance of various automated segmentation techniques.

Main Methods:

• Development of a novel multisequence MRI dataset for Wilms' tumor, including pre- and post-chemotherapy scans.
• Consensus-based ground truth annotation by five expert radiologists.
• Systematic evaluation of six segmentation algorithms, encompassing traditional and deep learning approaches.

Main Results:

• The created benchmark dataset provides a standardized resource for segmentation evaluation.
• Significant interrater variability was observed in manual annotations, particularly post-chemotherapy.
• The top-performing automated methods achieved segmentation quality comparable to human experts.

Conclusions:

• The developed benchmark dataset addresses a critical need in Wilms' tumor research.
• Automated segmentation tools show promise for improving the accuracy and consistency of tumor volume assessment.
• Further development of AI-driven segmentation could enhance pediatric cancer treatment planning.