Creation of a prototype simulator object for quality control of bone segmentation algorithms for use in tomographic post-processing software

Abstract

Considering the growing influence of Artificial Intelligence (AI) in medicine, especially in radiology, and the advances in computational power and theoretical understanding of AI algorithms (MESKÓ; GÖRÖG, 2020), concerns about the reliability of these methods arise. This study developed a prototype simulator object for quality control of bone segmentation algorithms in computed tomography. To achieve this objective, a 3D simulator object was created using advanced printing techniques, with materials that faithfully replicate the density and hardness of human bones, such as RadioMatrix. The results indicated that bone segmentation in post-processing software is strongly influenced by tissue density and that currently available AI models lack explainability (EXAI). The analysis revealed the need for algorithms that consider not only density but also the specific shapes and characteristics of bones. It is concluded that the developed prototype is effective in the evaluation and quality control of bone segmentation algorithms, promoting greater safety in clinical application and better explainability of bone segmentation algorithms in post-processing software for computed tomography.