Development of skin disease classifier using artificial intelligence

• Project/Area Number: 18K08290
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 53050:Dermatology-related
• Research Institution: University of Tsukuba
• Principal Investigator: Fujisawa Yasuhiro 筑波大学, 医学医療系, 准教授 (70550193)
• Project Period (FY): 2018-04-01 – 2021-03-31
• Project Status: Completed (Fiscal Year 2020)
• Budget Amount: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000); Fiscal Year 2020: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000); Fiscal Year 2019: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000); Fiscal Year 2018: ¥2,210,000 (Direct Cost: ¥1,700,000、Indirect Cost: ¥510,000)
• Keywords: 皮膚腫瘍 / 人工知能 / 分類 / ディープラーニング / 畳み込みニューラルネットワーク / 画像分類

Outline of Final Research Achievements

We trained AI using 4,800 out of 6,000 tumor images including benign and malignant which were retrospectively collected at University of Tsukuba, Dermatology division. The rest 1,200 images were used for the testing AI to calculate the accuracy of AI. Next, we randomly chose 140 images for the test dataset and compared the efficacy with board-certified dermatologists. As a result, AI classified images 93.4% accuracy whereas the board certified dermatologists achieved 85.3%, which was statistically significant different. Our study showed that the AI could classify tumor images more accurately than board-certified dermatologists.

Academic Significance and Societal Importance of the Research Achievements

本研究にて従来必要であるとされた1分類あたり1000枚という教師画像の数が，実際にはより少ない数（本研究では14クラスの分類で4800枚の教師画像）でも充分な精度で疾患の写真を分類できることを明らかにした．これは稀少な腫瘍が多い皮膚腫瘍の分野では多くの教師画像が集められないため，今後のAI皮膚腫瘍分類システムの構築にあたり大きなアドバンテージとなる．また，今回研究に使用した皮膚腫瘍だけでなくその他の分野にも応用が可能な技術であり，皮膚科に留まらずその他の分野においてもその開発の役に立つと思われる．現在はこのシステムを医療機器として使えるように研究開発を進めている．

Research Products

• [Journal Article] Deep‐learning‐based, computer‐aided classifier developed with a small dataset of clinical images surpasses board‐certified dermatologists in skin tumour diagnosis (2019)
  • Author(s): Fujisawa Y.、Otomo Y.、Ogata Y.、Nakamura Y.、Fujita R.、Ishitsuka Y.、Watanabe R.、Okiyama N.、Ohara K.、Fujimoto M.
  • Journal Title: British Journal of Dermatology Volume: 180 Issue: 2 Pages: 373-381
  • DOI: 10.1111/bjd.16924
  • Peer Reviewed
• [Journal Article] The Possibility of Deep Learning-Based, Computer-Aided Skin Tumor Classifiers (2019)
  • Author(s): Fujisawa Yasuhiro、Inoue Sae、Nakamura Yoshiyuki
  • Journal Title: Frontiers in Medicine Volume: 6 Pages: 00-00
  • DOI: 10.3389/fmed.2019.00191
  • NAID: 120007165449
  • Peer Reviewed
• [Journal Article] Melanoma and Non-Melanoma Skin Cancersメラノーマ・皮膚癌 人工知能による皮膚腫瘍の診断補助システム (2019)
  • Author(s): 藤澤 康弘
  • Journal Title: 癌と化学療法 Volume: 46 Pages: 637-640
• [Journal Article] 人工知能による皮膚腫瘍診断補助システムの開発 (2019)
  • Author(s): 藤澤 康弘，藤本 学，大友 雄造，藤田 亮
  • Journal Title: 画像ラボ Volume: 30 Pages: 30-35
• [Journal Article] 新しい検査法と診断法 人工知能(AI)による悪性腫瘍の診断 (2019)
  • Author(s): 藤澤 康弘, 藤本 学
  • Journal Title: 臨床皮膚科 Volume: 73 Pages: 64-68
• [Presentation] Can deep-learning-based, computer-aided classifier surpass board-certified dermatologists in skin tumour diagnosis? (2019)
  • Author(s): Fujisawa Y
  • Organizer: The 34th congress of Asia-Pacific Academy of Ophthalmology
  • Int'l Joint Research / Invited
• [Presentation] Computer-aided classifier with a deep convolutional neural network surpasses board-certified dermatologists in skin tumor diagnosis (2019)
  • Author(s): Fujisawa Y
  • Organizer: 15th European Association of Dermato-Oncology
  • Int'l Joint Research
• [Presentation] Can deep-learning-based, computer-aided classifier surpass board-certified dermatologistsin skin tumour diagnosis? (2019)
  • Author(s): Fujisawa Y
  • Organizer: The 34th congress of Asia-Pacific Academy of Ophthalmology
  • Int'l Joint Research / Invited