[22nd AIP Open Seminar] Talks by Machine Intelligence for Medical Engineering Team

Machine Intelligence for Medical Engineering Team (https://aip.riken.jp/labs/goalorient_tech/machine_intell_med_eng/) at RIKEN AIP

Speaker 1: Tatsuya Harada
Title: Overview of Machine Intelligence for Medical Engineering Team
Abstract:
We will present an overview of the Machine Intelligence for Medical Engineering Team. Medical information processing requires handling multimodal information such as 3D volumes, medical records with various constraints, and privacy-aware images. To tackle these topics, we have developed fundamental ML-based methods for 3D information processing, deep neural networks for tree structures, and privacy-aware knowledge transfer. We will also introduce those methods in this talk.

Speaker 2: Yusuke Mukuta
Title: Invariant Feature Coding using Tensor Product Representation
Abstract:
Exploiting the invariance is important for the efficient feature learning. We propose a method that incorporates the invariance into the image feature coding, which is the method to use the statistics of the local features as the global image feature. To this end, we regard the existing feature coding function as the tensor product of the local feature functions and then calculate its invariance as the global feature. The proposed method demonstrates better classification accuracy with robustness to the considered transformation using the smaller global feature dimension on several image recognition datasets including medical image analysis.

Speaker 3: Lin Gu
Title: Limited Data and Interpretability：the challenge and solution for Real-world Medical AI
Abstract:
Though deep learning has shown successful performance in the medical image analysis in the tasks of classifying the label and severity stage of certain disease. However, the CNN based methods suffer the bottleneck of lacking training label and interpretability. For example, most of them give few evidence on how to make prediction. To make it worse, the ubiquitous adversarial attack has posed even more serious challenge on its real application. This talk would introduce our recent progress for these challenges on various medical image domains for real-world application.

Speaker 4: Yusuke Kurose
Title: Machine Learning for Medical Image Diagnosis
Abstract:
The development of machine learning in recent years has been remarkable and has influenced many fields, and it is no exception in medical image processing. However, there are unique problems with medical image processing. For example, a very high resolution on a pathological image. It is not able to be applied to a general segmentation method directly due to the very high resolution on it. To solve this problem, a general method for pathological tissue classification uses a small patch extracted from the image as a classification input. However, it cannot consider a global feature in the tissue for classification. In this presentation, I will introduce our pathological segmentation method which can consider the global feature for the very high-resolution image. Also, I introduce other problems on medical image processing and our developed methods to solve them.