Details

Autor(en) / Beteiligte

• Guan, Shuyue
• Samala, Ravi K.
• Kahaki, Seyed M. M.
• Chen, Weijie

Titel

Restorable Synthesis: Average Synthetic Segmentation Converges to a Polygon Approximation of an Object Contour in Medical Images

Ist Teil von

• 2024 IEEE Southwest Symposium on Image Analysis and Interpretation (SSIAI), 2024, p.77-80

Ort / Verlag

IEEE

Erscheinungsjahr

2024

Quelle

IEEE Electronic Library (IEL)

Beschreibungen/Notizen

• Synthesis of segmentation contours is useful in evaluating truthing methods, i.e., the establishment of a segmentation reference standard by combining multiple segmentation results (e.g., by multiple experts). In contrast to a real-world application where the ground truth is often not available, the ground truth of objects is defined in synthetic data. Contours with combinations of segmentation errors, as compared to the defined ground truth, can be synthesized. A desired property of segmentation contour synthesis for evaluating truthing methods, which we call the restorability property, is that the average of multiple segmentation contours can converge to the truth contour. This property is desired because such a dataset can serve as a benchmark for evaluating if commonly used truthing methods have bias. We developed a segmentation contour synthesis tool that has the restorability property and conducted simulation studies to validate this tool.