Development of cytological diagnosis aiding platform

2020 Fiscal Year Final Research Report

• Project/Area Number: 18K15310
• 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: University of Yamanashi
• Principal Investigator: Kawai Masataka 山梨大学, 大学院総合研究部, 臨床助教 (00813239)
• Project Period (FY): 2018-04-01 – 2021-03-31
• Keywords: 細胞診 / 甲状腺癌 / 深層学習

Outline of Final Research Achievements

We collected cytological images and trained deep neural networks. Seven papillary thyroid carcinoma (PTC) and six non-PTC cases were collected from University of Yamanashi Hospital. Cytologic slides were scanned with virtual slide scanner (VS-120, Olympus). Deep neural models were implemented with Keras (TensorFlow backend). We trained VGG16, GoogLeNet, Xception, InceptionV3, ResNet50, and original model (2 convolution+batchnormalization followed by 2 layer multi-layer perceptron) as PTC or non-PTC classification task. Fine-tuned ResNet50 pretrained on ImageNet resulted in as high as 98% accuracy. We are investigating generative adversarial network (GAN) to transfer histological images knowledge to cytological diagnosis.

Free Research Field

腫瘍学

Academic Significance and Societal Importance of the Research Achievements

細胞診のスクリーニングや診断は人材の不足や検体の増加に伴い、自動化が望まれている。本研究により細胞診画像への深層学習応用が有用であることが示され、細胞診へのAI導入が有用であることが示唆された。