Keywords:
Artificial Intelligence, Computer applications, Musculoskeletal system, MR, Computer Applications-3D, Computer Applications-General, Imaging sequences, Athletic injuries
Authors:
S. Ramedani, H. Von Tengg-Kobligk, C. Morhard, K. Daneshvar Ghorbani
DOI:
10.26044/ecr2022/C-20511
Purpose
Several studies have shown that inappropriate amounts of fat can considerably raise the risk of many diseases. MSK disorders constitute the world's second leading cause of disability based on years lived with disability (YLDs) [1]. It is estimated that disability resulting from MSK diseases, specifically osteoarthritis (OA), has increased by 45% between 1990 and 2010 and will further grow with an increasingly obese, inactive and ageing population [2]. Thus, accurate and precise measurements of muscle and fat volumes are essential for a better understanding of various diseases, syndromes and disorders. MRI as an imaging method can assess the volume of body components. The main benefit of MRI in comparison to other techniques (e.g., DEXA Scan) is the ability to precisely map the regional body composition without ionizing radiation. The latest developments in fat and water discrimination (e.g., Dixon sequence) using 3D multi-echo gradients have further improved soft tissue contrast and the measurement accuracy of fat infiltration in skeletal muscle achievable by MRI [3]. The Dixon method supplies up to four different contrasts in the one acquisition. However, most publications have not used the combined image of Dixon technique, instead they used one fat or water image after another to segment the fat or muscle elements individually.
Manual segmentation is tedious, time-consuming, and inconvenient for large-scale studies. Therefore, numerous studies have been done to provide multiple automatic segmentation methods for MRI images recently. In recent years, with the appearance of machine learning approaches and deep neural networks, medical imaging is taking advantage of these approaches as well. But despite the significant advances in automatic segmentation and quantitative analysis of medical images that have been presented in the literature, no explicit automatic whole-body MRI segmentation framework is available which uses all four sequences of the Dixon technique in order to accurately segment fat and muscle, while also segmenting muscle types.
The aim of this study is to develop a novel machine learning based and fully automated muscle-fat body composition assessment framework that completely leverages the combination of all four sequences generated by the Dixon technique. The hypothesis of this study is that the problem of limited data can be addressed with novel artificial intelligences architectures.