Realization of Multidisciplinary Understanding of Skeletal Muscle by Constructing Deep Learning and Model Integration Theory

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K12731
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 90130:Medical systems-related
• Research Institution: Aichi Prefectural University
• Principal Investigator: Kamiya Naoki 愛知県立大学, 情報科学部, 准教授 (00580945)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 骨格筋 / 脊柱起立筋 / L3断面 / 体組成 / 体腔 / 深層学習 / セグメンテーション

Outline of Final Research Achievements

In this research project, we aimed to develop a skeletal muscle segmentation technology that overcomes the challenges of whole-body skeletal muscle recognition by employing deep learning and model integration theory, as opposed to conventional single-element skeletal muscle recognition and analysis techniques based on skeletal muscle recognition. As a baseline, we used the automatic recognition of 8 relatively shape-specific skeletal muscles, which were previously recognized by model-based descriptions of muscle contours. We then developed a deep learning and model-based recognition technology that can be expanded to whole-body skeletal muscle recognition, compatible with annotation costs. This includes a recognition method for region-specific skeletal muscles using the erector spinae and body cavity as keys, as well as a real-quantity measurement technology for whole-body body composition, including skeletal muscle, rather than mere estimation.

Free Research Field

医用画像情報処理

Academic Significance and Societal Importance of the Research Achievements

本研究は，骨格筋の自動認識と解析を新しい視点で再定義することを目的として実施した．骨格筋の解析は医師による半自動セグメンテーションが主流であり，主に横断面積に焦点を当てているが，我々は深層学習ベースの骨格筋認識と形状・筋束モデルの併用による，全身規模による認識を目指した．本研究により，深層学習とモデルを併用することで，従来の単一要素による解析にとどまらず，筋のミクロ・マクロ構造の多角的な解析の糸口を構築できた．本研究は，萎縮性筋疾患の画像鑑別など新たな医学的見地を得るために必要な要素技術の一つとしての可能性があり，医学的，工学的，産業的にも意義を持つ．