Abstract: SA-PO726
Artificial Intelligence Is Useful for Quantitative Analyses of Human Kidney Biopsy Images
Session Information
- Pathology and Lab Medicine: Clinical
November 09, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Pathology and Lab Medicine
- 1602 Pathology and Lab Medicine: Clinical
Authors
- Matsumoto, Ayumi, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Matsui, Isao, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Shimada, Karin, Graduate School of Medicine, Osaka University, Osaka, Osaka-Fu, Japan
- Hashimoto, Nobuhiro, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Yasuda, Seiichi, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Kajimoto, Sachio, Osaka university, Osaka, Japan
- Doi, Yohei, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Yamaguchi, Satoshi, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Oka, Tatsufumi, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Kubota, Keiichi, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Sakaguchi, Yusuke, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Namba, Tomoko, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Inoue, Kazunori, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Hamano, Takayuki, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
- Isaka, Yoshitaka, Osaka University Graduate School of Medicine, Suita, OSAKA-FU, Japan
Background
The kidney biopsy is a valuable clinical examination that provides profound insight into the diagnosis, prognosis, and treatment of kidney diseases. However, appropriate quantitative analyses of the kidney biopsy images require trained nephrologists and pathologists. Artificial intelligence, an emerging technology in the field of computer science, may help analyses of pathology images.
Methods
PAS, PAM, or Masson-stained human kidney biopsy samples were obtained from four hospitals in Japan. The kidney biopsy images were digitalized by virtual slide system. Convolutional neural network (CNN), a deep neural network for computer vision, was trained to segment the kidney biopsy images.
Results
The CNN was almost well trained to identify glomerular and global sclerosis regions in various kidney diseases. However, Kimmelstiel-Wilson nodules were tended to be identified as global sclerosis. The CNN could also segment kidney biopsy images into intact tubulointerstitium, mild expansion of interstitial matrix, severe expansion of interstitial matrix, inflammation, arteries, and capsule of the kidney. Because PAS, PAM, and Masson-staining techniques were slightly different among the four hospitals, the training of the CNN required data from all four hospitals to obtain better accuracy of segmentation.
Conclusion
The artificial intelligence could segment human kidney biopsy images. The segmentation by artificial intelligence can provide novel quantitative methods for the analyses of human kidney biopsy images.
Segmentation of kidney biopsy images by artificial intelligence