Construction of an Ultra-early Nail Melanoma Diagnosis System Using Artificial Intelligence with Deep Learning

2020 Fiscal Year Final Research Report

• Project/Area Number: 18K15271
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 50020:Tumor diagnostics and therapeutics-related
• Research Institution: Shinshu University
• Principal Investigator: koga hiroshi 信州大学, 学術研究院医学系(医学部附属病院), 講師 (30419361)
• Project Period (FY): 2018-04-01 – 2021-03-31
• Keywords: 爪甲色素線条 / 人工知能 / メラノーマ

Outline of Final Research Achievements

Dermoscopy photographs of melanonychia visited at the outpatient clinic of Shinshu University Hospital between 2007 and 2020 were used as samples. Clinical information for each case was investigated and data cleaning was performed, followed by correct labeling or annotation for each case image. The image population was divided into training, validation, and verification datasets. The training image dataset was used to create an initial classifier using a general-purpose deep learning tool. We evaluated the classification performance on the validation data.

Free Research Field

皮膚科学

Academic Significance and Societal Importance of the Research Achievements

希少疾患である爪部メラノーマの初期症状である爪甲色素線条を、その他の原因で生じる爪甲色素線条と臨床的に区別するための補助となるプログラムを作成した。既に海外で報告されているAIプログラムの性能を上回ることができたが、それは主に組み入れ対象の調整による影響が大きいと考えられた。希少疾患のAI作成では学習データ不足が問題となり、全国的なデータ収集体制とともにAI分類機作成にあたっての技術的な工夫が必要である。