The main purpose of the new academic area research “Multidisciplinary Computational Anatomy” was multi-dimensionalization of “Computational Anatomy”, which is the result of the fusion of high-definition medical imaging technology and informatics. In the newly launched multidisciplinary computational anatomy, based on the research results of “Computational Anatomy”, we have established a multidisciplinary computational anatomy model with a spatial axis of analysis from micro at the cell level to macro at the organ level, a time axis of analysis of various periods for the human life from the development of the fetus to the death, a functional axis of analysis targeting physiological functions and metabolism of cells and organs, and a pathological axis of analysis from normal to pathological diseases such as inflammation and carcinogenic process. Then, we used this multi-dimensional computational anatomy model to fuse and integrate each axis to deepen the true understanding of the human body. Here, we introduce the outline of the many results of multidisciplinary computational anatomy, which have been applied to the diagnosis and treatment methods for diseases that are difficult to detect early and treat.

EndoBRAIN (Cybernet System Corp.; Olympus Corp., Tokyo) is the computer-aided diagnostic system supportedby artificial intelligence (AI) that allows real-time identification of colorectal polyps. EndoBRAIN was the firstly approved medical device in the field of AI in Japan. This regulatory approval was achieved by the close collaboration among medical researchers, engineering researchers, industrial partners, and public funding bodies. This review article shows the overview of EndoBRAIN including its design, performance, and results in clinical trials. In addition, our approaches to regulatory approval and insurance reimbursement, which unfortunately has not been granted, are also introduced.

This paper introduces our achievements of the researches in JSPS KAKENHI “Clinical Applications of Multidisciplinary Computational Anatomy to Diagnosis (Grant Number: 26108009)”. In this research, we developed methods for various applications. We introduce researches focusing on the computer-aided diagnosis (CAD) systems for lung tumors in this paper. Most CAD systems for lung tumor in chest CT images process a whole method in sequence of detection, segmentation, and discrimination. In this paper, we introduce a method to detect tumors by using temporal subtraction between the previous and current images, a method to segment tumors by using the current images and the temporal subtraction images, and a method to discriminate between benign and malignant tumors in the current images based on medical findings.

The goal of this study is to transform to the digitized intra-operative imaging and the compiled brain-function database for the predicting glioma surgery that is based on patientʼs future perspective depending on the tumor resection rate as well as the post-operative complication rate. In awake craniotomy, we estimated language-related location in response to the surgeonʼs electrical stimulation and the examinerʼs task from the precise process analysis of the medical device “IEMAS: Intra-operative examination monitoring in awake surgery”. Secondarily, successful acquisition of log data with the location of medical device integrated into intra-operative MR image was performed and digitized brain function was converted to a normalized brain data format. Digitized log data of the electrostimulation probe during awake craniotomy was acquired successfully in 20 cases, that were totally 22 speech arrest (SA), 10 speech impairment (SI), 12 motor, and 7 sensory responses (51 responses). Finally, intraoperative SA response converted fully to normalized brain with acceptable accuracy. We simulated the projection of the normalized brain data to the individual pre- and intra-operative MR image. These image integration and transformation methods using brain normalization should facilitate practical intra-operative brain mapping. These methods may be helpful for pre-operatively and/or intra-operatively predicting brain function.

Endoscopic surgery is an interaction between surgical instruments and organs performed by an operating surgeon based on endoscopic images. Image analysis technology that extracts quantitative intraoperative information in real time is paramount for improving the levels of autonomy (LoAs) of surgical assistant robots supporting the surgeon. This paper reports two achievements of the research group A03-KB102―entitled “Autonomous control of a surgical assistant robot with a stereoscopic endoscope that can obtain depth information in real time”―in the multidisciplinary computational anatomy (MCA) project: (1) estimating the distance between the tip of a surgical instrument and the organ surface and (2) estimating the load of the surgical instrumentʼs tip during counter traction against an organ. The stereo matching engine developed in this study can offer a new modality in the construction of an MCA model. The integration of our system and other MCA modalities is promising for further improvement of LoAs in robotic surgery.

We previously proposed a pulmonary nodule clarification method for chest radiographs that controlled for the pulmonary vessels that were frequently extracted as false positives. In addition, further true pulmonary nodules were detected by applying a wavelet analysis and an error diffusion method to control for density alterations caused by clavicles, ribs, and peripheral pulmonary vessel shadows (background noise). However, this method was insufficient for extracting pulmonary nodules at the level of the pulmonary hilum. We herein report a new method for detecting such pulmonary nodules by applying cellular automata and adaptive rank filtering to the binary image produced using the error diffusion method. Two radiologists compared the new images obtained by the proposed technique with the background noise-suppressed pulmonary nodule-clarified images regarding suppression of the background noise and visibility (degree of emphasis) of the pulmonary nodules. This evaluation used 117 images with pulmonary nodules from the Japan Society of Radiological Technology database, excluding “extremely subtle” and “obvious” pulmonary nodules. While the pulmonary nodules at the level of the pulmonary hilum were enhanced, the background noise using the proposed method was not higher than that in our previous method in 76.1% of cases. The visibility of the pulmonary nodules was improved in 12.8% of cases. The proposed method for clarifying pulmonary nodules is expected to improve the detection of lung cancer nodules.