Development of methods for automated preprocedural CT assessment for TAVR by artificial intelligence

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K20920
• Research Category: Grant-in-Aid for Research Activity Start-up
• Allocation Type: Multi-year Fund
• Review Section: 0903:Organ-based internal medicine and related fields
• Research Institution: Jikei University School of Medicine
• Principal Investigator: Miyasaka Masaki 東京慈恵会医科大学, 医学部, 助教 (60911154)
• Project Period (FY): 2021-08-30 – 2023-03-31
• Keywords: AI / TAVI / TAVR / CT / landmark detection / 3D / segmentation

Outline of Final Research Achievements

This study aimed to establish a foundation for the application of artificial intelligence (AI) to three-dimensional imaging modalities by applying AI to cardiac CT analysis for preoperative evaluation of Transcatheter Aortic Valve Implantation (TAVI). During development, the step of accurately estimating planes from three-dimensional data was highly valuable due to its broad applicability but also posed significant challenges.
By developing an algorithm based on the optimal transport distance, we achieved high-precision plane estimation. This method allows for the localization of landmarks with accuracy comparable to CT image analysis experts, even from low-resolution CT images, while yielding higher accuracy compared to existing AI technologies. A manuscript on these findings for publication has being prepared.

Free Research Field

Cardiology

Academic Significance and Societal Importance of the Research Achievements

2次元画像に比べて3次元画像のAI開発は難しいとされるが、本研究は同分野に進展をもたらした。本研究で提唱したUNET-GliPにより、ヒートマップ回帰における学習の安定性と予測精度のトレードオフを緩和し、粗い解像度の1.6mm立方ボクセルのCT画像からランドマークを高精度で検出することに成功した。この技術は、TAVI術前CT解析にとどまらず、僧帽弁逆流症などの他の構造的心疾患の診断、手術プランニング、さらにMRIや心臓超音波検査などの異なる3次元画像モダリティへの応用基盤としての可能性を秘めており、診断精度向上、診療負担軽減、医用画像解析研究の発展に貢献することが期待される。