Automated analysis of whole slide image (WSI)

The automated WSI analysis system is applied to histopathology diagnosis of hepatocellular carcinoma [Fig. 1]. The overview of the system is shown in fig. 2. After preprocessing such as color correction, the nuclei detection and nuclear feature measurement are employed. Also, stromal areas, sinusoids and fat droplets are segmented for the structural feature measurements. The system shows the quantification results as a heat map. Moreover, the cancer discrimination is performed by SVM (support vector machine) using the feature vector composed of nuclei and structural features. The result of feature measurement and SVM classification are also visualized as a heat map as shown in fig. 3. The accuracy of classification was tested by 5-fold cross-validation using about 1,000 ROIs. As a result, both the sensitivity and specificity were almost 90% [Table 1]. The heat map of the possibility of malignancy will support pathologists to pick up the regions that should be observed carefully, and the heat map of each feature provides pathologists supportive information for subtype classification, grading, treatment selection, and reasoning of the diagnosis.
Demonstration at IADP 2014 / CIC22 (2014, Boston)
[Link https://www.oid.ict.e.titech.ac.jp/wp/home_en/research/pathology/automated-analysis/iadp-2014-cic22/ ]

Fig. 1 Prototype system

Fig. 2 Flow diagram of the prototype system

Fig. 3 Heat map for HCC probability

Table 1 Classification result

The research project, “image analysis technology for quantitative pathological diagnosis” was carried out by Keio University, NEC Corporation, Saitama Medical University and Tokyo Institute of Technology as a commissioned research by NEDO (New Energy and Industrial Technology Development Organization). I acknowledge Professor Michiie Sakamoto, Dr. Akinori Hashiguchi, and Dr. Tokiya Abe, Keio University, Dr. Akira Saito, Dr. Kenichi Kamijo, Dr. Tomoharu Kiyuna, Kamei, Dr. Yoshiko Yamashita, NEC, Professor Naoki Kobayashi and Dr. Masahiro Ishikawa, Saitama Medical School, Dr. Yuri Murakami and Professor Hiroshi Nagahashi, formerly Tokyo Institute of Technology, Dr. Hiroshi Kanazawa, Dr. Fumio Kimura, and the students in Optical Imaging and Display group, Tokyo Institute of Technology.

Automated analysis of whole slide image (WSI)

References