Cloud-Based Evaluation of Anatomical Structure Segmentation and Landmark Detection Algorithms: VISCERAL Anatomy...

Area of Science:

• Medical Imaging Analysis
• Computational Anatomy
• Radiology

Background:

• Anatomical variations in medical images are key radiological indicators of disease.
• Manual analysis of these variations is time-consuming and requires expert knowledge.
• Automated tools are needed to streamline the process of analyzing anatomical structures.

Purpose of the Study:

• To present the VISCERAL Anatomy benchmarks, a cloud-based framework for evaluating medical imaging algorithms.
• To benchmark state-of-the-art algorithms for anatomical structure segmentation and landmark detection.
• To provide an objective comparison of algorithm performance across multiple imaging modalities.

Main Methods:

• Development of a cloud-based evaluation framework utilizing virtual machines.
• Manual annotation of 120 computed tomography (CT) and magnetic resonance (MR) patient volumes, creating a Gold Corpus.
• Benchmarking of 10 organ segmentation algorithms and 3 landmark localization algorithms on a common, unseen test set.

Main Results:

• The VISCERAL Anatomy benchmarks facilitated objective performance comparison of automated algorithms.
• Different algorithms achieved top scores across various imaging modalities and anatomical structures.
• The study highlights the variability in performance of current state-of-the-art algorithms.

Conclusions:

• The VISCERAL Anatomy benchmarks provide a robust framework and valuable dataset for advancing medical image analysis.
• The results offer insights into the strengths and weaknesses of current segmentation and landmark detection algorithms.
• The VISCERAL dataset and Silver Corpus are publicly available to the research community for further development.