The starred authors(*) are co-first authors and contributed equally.

Development of Frozen Tuna Inspection Device using Ultrasound and Machine Learning


    In progress
    1. Akira Sakai, Masafumi Yagi, Suguru Yasutomi, Kazuki Mizuno, Kanata Suzuki, Keiichi Goto: Machine Learning Approach for Frozen Tuna Freshness Inspection using Low-frequency A-mode Ultrasound, IEEE Access, vol.11, pp.107379-107393, 2023.

The Study of Life-Long Machine Learning using Daily Necessities Image Dataset (DAISO-100)


    In machine learning, especially deep learning, collecting high-quality data has a significant impact on model performance. We attempted to establish an efficient learning method by utilizing information linked to data domains. In [1, 2], we collected weakly labeled data by using a robotic arm to capture sequential images around an object. Afterwards, by performing active learning [2] or metric learning [1] while evaluating each piece of data, efficient data collection and model training were achieved. The constructed dataset is publicly available. Furthermore, in [3], we used the dataset collected in [2] to verify learning methods (late-stopping, tuning the momentum parameter of the batch normalization, enforcing invariance of the neural activity) to improve the recognition accuracy of out-of-distribution (OoD) data. We demonstrated that even though the three approaches focus on different aspects of DNNs, they all tend to lead to the same underlying neural mechanism to enable OoD accuracy gains — individual neurons in the intermediate layers become invariant to OoD orientations and illuminations.
  1. Kanata Suzuki, Yasuto Yokota, Yuzi Kanazawa, Tomoyoshi Takebayashi: Online Self-Supervised Learning for Object Picking: Detecting Optimum Grasping Position using a Metric Learning Approach, Proceedings of 2020 IEEE/SICE International Symposium on System Integrations (SII'20), pp.205-212, Honolulu, USA, January 12-15, 2020.
  2. Kanata Suzuki, Taro Sunagawa, Tomotake Sasaki, Takashi Katoh: Annotation Cost Reduction of Stream-based Active Learning by Automated Weak Labeling using a Robot Arm, Proceedings of 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS'21), pp.9000-9007, acceptance rate 45%, Online, September 27- October 1, 2021, [Dataset] [Blog].
  3. Akira Sakai, Taro Sunagawa, Spandan Madan, Kanata Suzuki, Takashi Katoh, Hiromichi Kobashi, Hanspeter Pfister, Pawan Sinha, Xavier Boix, Tomotake Sasaki: Three approaches to facilitate DNN generalization to objects in out-of-distribution orientations and illuminations, Neural Networks, vol.155, pp.119-143, 2022.

Medical Image Analysis using Deep Neural Network


    With the recent development of deep learning technology, its application to medical image analysis is progressing. However, medical images are sensitive data that includes personal information, and there are barriers to practical use in diagnosis due to data collection and prediction accuracy issues. In this study, we focused on the detection of malignant cases in medical diagnosis, especially the detection of glomeruli in kidney images and the detection of finger joints in X-ray images of limbs. In [1, 2], by introducing an AI system into electronic medical records, it became possible to efficiently annotate highly confidential data during the work of medical professionals. In addition, [3] compared the diagnostic results of experienced clinicians and confirmed that diagnostic accuracy was improved by using a majority voting method that combined human diagnosis and AI model prediction.
    1. Keisuke Izumi*, Kanata Suzuki*, Masahiro Hashimoto, Masahiro Jinzaki, Shigeru Ko, Tsutomu Takeuchi, Yuko Kaneko: Ensemble Detection of Hand Joint Ankylosis and Subluxation in Radiographic Images using Deep Neural Networks, Scientific Reports, 14, 7696, 2024.
    2. Keisuke Izumi*, Kanata Suzuki*, Masahiro Hashimoto, Toshio Endoh, Kentaro Doi, Yuki Iwai, Masahiro Jinzaki, Shigeru Ko, Tsutomu Takeuchi, Yuko Kaneko: Detecting hand joint ankylosis and subluxation in radiographic images using deep learning: A step in the development of an automatic radiographic scoring system for joint destruction, PLOS ONE,  18(2): e0281088, 2023.
    3. Eiichiro Uchino*, Kanata Suzuki*, Noriaki Sato, Ryosuke Kojima, Yoshinori Tamada, Shusuke Hiragi, Hideki Yokoi, Nobuhiro Yugami, Sachiko Minamiguchi, Hironori Haga, Motoko Yanagita, Yasushi Okuno: Classification of glomerular pathological findings using deep learning and nephrologist–AI collective intelligence approach, International Journal of Medical Informatics, vol. 141(2020), 104231, 2020.