Medical image understanding based on multimodal deep representation learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K12077
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61050:Intelligent robotics-related
• Research Institution: Chukyo University
• Principal Investigator: Mekada Yoshito 中京大学, 工学部, 教授 (00282377)
• Co-Investigator(Kenkyū-buntansha): 道満 恵介 中京大学, 工学部, 准教授 (90645748)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 深層学習 / 医用画像処理 / 特徴選択

Outline of Final Research Achievements

The outcomes of this research project are to improve the classification accuracy through machine learning and deep learning by appropriately combining both low-dimensional but useful features, such as inspection data, and redundant yet high-dimensional features, such as images rich in morphological information. By transforming image features into comprehensive intermediate features, like Radiomics features, and reducing their dimensionality while further combining feature selection methods to match the dimensionality of the inspection data, we demonstrated that classification accuracy could be improved. Additionally, to enhance the interpretability of results such as lesion detection, we developed a two-stage estimation method for predicting a standardized method using 25 abdominal ultrasound images. The results showed an accuracy rate of 83.6%.

Free Research Field

パターン認識

Academic Significance and Societal Importance of the Research Achievements

分子標的治療薬の病勢制御については、分子標的治療薬レンバチニブの薬事承認のためのデータの一部を用いたが、これらの治療方法の選択に時間をかける余裕はなく、今回実現した効果予測手法に対して、さらなる検証実験を行うことで治療成績の向上が期待できる。また、撮影と診断を同時に行わなければならない超音波スクリーニングにおいて、標準的な断面を推定できることでスクリーニング時の診断精度の向上が期待できる。