Isomorphic Mesh Generation from Point Clouds with Multilayer Perceptrons.

A novel neural network called the isomorphic mesh generator (iMG) is proposed to generate isomorphic meshes from point clouds containing noise and missing parts. Isomorphic meshes of arbitrary objects exhibit a unified mesh structure, despite objects belonging to different classes. This unified representation enables various modern deep neural networks (DNNs) to easily handle surface models without requiring additional pre-processing. Additionally, the unified mesh structure of isomorphic meshes enables the application of the same process to all isomorphic meshes, unlike general mesh models, where processes need to be tailored depending on their mesh structures. Therefore, the use of isomorphic meshes can ensure efficient memory usage and reduce calculation time. Apart from the point cloud of the target object used as input for the iMG, point clouds and mesh models need not be prepared in advance as training data because the iMG is a data-free method. Furthermore, the iMG outputs an isomorphic mesh obtained by mapping a reference mesh to a given input point cloud. To stably estimate the mapping function, a step-by-step mapping strategy is introduced. This strategy enables flexible deformation while simultaneously maintaining the structure of the reference mesh. Simulations and experiments conducted using a mobile phone have confirmed that the iMG reliably generates isomorphic meshes of given objects, even when the input point cloud includes noise and missing parts.

Personalized Federated Learning for Institutional Prediction Model using Electronic Health Records: A Covariate Adjustment Approach.

Federated learning (FL) has attracted attention as a technology that allows multiple medical institutions to collaborate on AI without disclosing each other's patient data. However, FL has the challenge of being unable to robustly learn when the data of participating clients is non-independently and non-identically distributed (Non-IID). Personalized Federated Learning (PFL), which constructs a personalized model for each client, has been proposed as a solution to this problem. However, conventional PFL methods do not ensure the interpretability of personalization, specifically, the identification of which data samples are contributed to each personalized learning, which is important for AI in medical applications. In this study, we propose a novel PFL framework, Federated Adjustment of Covariate (FedCov), which acquires a propensity score model representing the covariate shift among clients through prior FL, then learns a final model by weighting the contribution of each training sample to PFL based on the estimated propensity score. This approach enables both the learning of personalized models through covariate adjustment and the visualization of the contribution of each client to PFL. FedCov was evaluated in the prediction of in-hospital mortality across 50 hospitals in the eICU Collaborative Research Database, achieving an ROC-AUC of 0.750. This result outperformed the AUCs in the 0.720-0.735 range achieved by conventional FL methods and was closest to the AUC of 0.754 achieved by centralized learning.Clinical Relevance- This study demonstrates the feasibility of providing sophisticated and personalized AI-driven clinical decision support to any medical institution through personalized federated learning.

Augmented reality-based affective training for improving care communication skill and empathy.

It is important for caregivers of people with dementia (PwD) to have good patient communication skills as it has been known to reduce the behavioral and psychological symptoms of dementia (BPSD) of PwD as well as caregiver burnout. However, acquiring such skills often requires one-on-one affective training, which can be costly. In this study, we propose affective training using augmented reality (AR) for supporting the acquisition of such skills. The system uses see-through AR glasses and a nursing training doll to train the user in both practical nursing skills and affective skills such as eye contact and patient communication. The experiment was conducted with 38 nursing students. The participants were assigned to either the Doll group, which only used a doll for training, or the AR group, which used both a doll and the AR system. The results showed that eye contact significantly increased and the face-to-face distance and angle decreased in the AR group, while the Doll group had no significant difference. In addition, the empathy score of the AR group significantly increased after the training. Upon analyzing the correlation between personality and changes of physical skills, we found a significant positive correlation between the improvement rate of eye contact and extraversion in the AR group. These results demonstrated that affective training using AR is effective for improving caregivers' physical skills and their empathy for their patients. We believe that this system will be beneficial not only for dementia caregivers but for anyone looking to improve their general communication skills.

Artificial Intelligence for Segmentation of Bladder Tumor Cystoscopic Images Performed by U-Net with Dilated Convolution.

Background: Early intravesical recurrence after transurethral resection of bladder tumors (TURBT) is often caused by overlooking of tumors during TURBT. Although narrow-band imaging and photodynamic diagnosis were developed to detect more tumors than conventional white-light imaging, the accuracy of these systems has been subjective, along with poor reproducibility due to their dependence on the physician's experience and skills. To create an objective and reproducible diagnosing system, we aimed at assessing the utility of artificial intelligence (AI) with Dilated U-Net to reduce the risk of overlooked bladder tumors when compared with the conventional AI system, termed U-Net. Materials and Methods: We retrospectively obtained cystoscopic images by converting videos obtained from 120 patients who underwent TURBT into 1790 cystoscopic images. The Dilated U-Net, which is an extension of the conventional U-Net, analyzed these image datasets. The diagnostic accuracy of the Dilated U-Net and conventional U-Net were compared by using the following four measurements: pixel-wise sensitivity (PWSe); pixel-wise specificity (PWSp); pixel-wise positive predictive value (PWPPV), representing the AI diagnostic accuracy per pixel; and dice similarity coefficient (DSC), representing the overlap area between the bladder tumors in the ground truth images and segmentation maps. Results: The cystoscopic images were divided as follows, according to the pathological T-stage: 944, Ta; 412, T1; 329, T2; and 116, carcinoma in situ. The PWSe, PWSp, PWPPV, and DSC of the Dilated U-Net were 84.9%, 88.5%, 86.7%, and 83.0%, respectively, which had improved when compared to that with the conventional U-Net by 1.7%, 1.3%, 2.1%, and 2.3%, respectively. The DSC values were high for elevated lesions and low for flat lesions for both Dilated and conventional U-Net. Conclusions: Dilated U-Net, with higher DSC values than conventional U-Net, might reduce the risk of overlooking bladder tumors during cystoscopy and TURBT.

Classification of Motor Impairments of Post-Stroke Patients Based on Force Applied to a Handrail.

Many patients suffer from declined motor abilities after a brain injury. To provide appropriate rehabilitation programs and encourage motor-impaired patients to participate further in rehabilitation, sufficient and easy evaluation methodologies are necessary. This study is focused on the sit-to-stand motion of post-stroke patients because it is an important daily activity. Our previous study utilized muscle synergies (synchronized muscle activation) to classify the degree of motor impairment in patients and proposed appropriate rehabilitation methodologies. However, in our previous study, the patient was required to attach electromyography sensors to his/her body; thus, it was difficult to evaluate motor ability in daily circumstances. Here, we developed a handrail-type sensor that can measure the force applied to it. Using temporal features of the force data, the relationship between the degree of motor impairment and temporal features was clarified, and a classification model was developed using a random forest model to determine the degree of motor impairment in hemiplegic patients. The results show that hemiplegic patients with severe motor impairments tend to apply greater force to the handrail and use the handrail for a longer period. It was also determined that patients with severe motor impairments did not move forward while standing up, but relied more on the handrail to pull their upper body upward as compared to patients with moderate impairments. Furthermore, based on the developed classification model, patients were successfully classified as having severe or moderate impairments. The developed classification model can also detect long-term patient recovery. The handrail-type sensor does not require additional sensors on the patient's body and provides an easy evaluation methodology.

Speed invariant gait recognition-The enhanced mutual subspace method.

This paper introduces an enhanced MSM (Mutual Subspace Method) methodology for gait recognition, to provide robustness to variations in walking speed. The enhanced MSM (eMSM) methodology expands and adapts the MSM, commonly used for face recognition, which is a static/physiological biometric, to gait recognition, which is a dynamic/behavioral biometrics. To address the loss of accuracy during calculation of the covariance matrix in the PCA step of MSM, we use a 2D PCA-based mutual subspace. Furhtermore, to enhance the discrimination capability, we rotate images over a number of angles, which enables us to extract richer gait features to then be fused by a boosting method. The eMSM methodology is evaluated on existing data sets which provide variable walking speed, i.e. CASIA-C and OU-ISIR gait databases, and it is shown to outperform state-of-the art methods. While the enhancement to MSM discussed in this paper uses combinations of 2D-PCA, rotation, boosting, other combinations of operations may also be advantageous.

3D segmentation of nasopharyngeal carcinoma from CT images using cascade deep learning.

In the paper, we propose a new deep learning-based method for segmenting nasopharyngeal carcinoma (NPC) in the nasopharynx from three orthogonal CT images. The proposed method introduces a cascade strategy composed of two-phase manners. In CT images, there are organs, called non-target organs, which NPC never invades. Therefore, the first phase is to detect and eliminate non-target organ regions from the CT images. In the second phase, NPC is extracted from the remained regions in the CT images. Convolutional neural networks (CNNs) are applied to detect non-target organs and NPCs. The proposed system determines the final NPC segmentation by integrating three results obtained from coronal, axial and sagittal images. Moreover, we construct two CNN-based NPC detection systems using one kind of overlapping patches with a fixed size and various overlapping patches with different sizes. From the experiments using CT images of 70 NPC patients, our proposed systems, especially the system using various patches, achieves the best performance for detecting NPC compared with conventional NPC detection methods.

Fast modified Self-organizing Deformable Model: Geometrical feature-preserving mapping of organ models onto target surfaces with various shapes and topologies.

BACKGROUND AND OBJECTIVE: This paper proposes a new method for mapping surface models of human organs onto target surfaces with the same genus as the organs. METHODS: In the proposed method, called modified Self-organizing Deformable Model (mSDM), the mapping problem is formulated as the minimization of an objective function which is defined as the weighted linear combination of four energy functions: model fitness, foldover-free, landmark mapping accuracy, and geometrical feature preservation. Further, we extend mSDM to speed up its processes, and call it Fast mSDM. RESULTS: From the mapping results of various organ models with different number of holes, it is observed that Fast mSDM can map the organ models onto their target surfaces efficiently and stably without foldovers while preserving geometrical features. CONCLUSIONS: Fast mSDM can map the organ model onto the target surface efficiently and stably, and is applicable to medical applications including Statistical Shape Model.

An informationally structured room for robotic assistance.

The application of assistive technologies for elderly people is one of the most promising and interesting scenarios for intelligent technologies in the present and near future. Moreover, the improvement of the quality of life for the elderly is one of the first priorities in modern countries and societies. In this work, we present an informationally structured room that is aimed at supporting the daily life activities of elderly people. This room integrates different sensor modalities in a natural and non-invasive way inside the environment. The information gathered by the sensors is processed and sent to a centralized management system, which makes it available to a service robot assisting the people. One important restriction of our intelligent room is reducing as much as possible any interference with daily activities. Finally, this paper presents several experiments and situations using our intelligent environment in cooperation with our service robot.

Floor sensing system using laser reflectivity for localizing everyday objects and robot.

This paper describes a new method of measuring the position of everyday objects and a robot on the floor using distance and reflectance acquired by laser range finder (LRF). The information obtained by this method is important for a service robot working in a human daily life environment. Our method uses only one LRF together with a mirror installed on the wall. Moreover, since the area of sensing is limited to a LRF scanning plane parallel to the floor and just a few centimeters above the floor, the scanning covers the whole room with minimal invasion of privacy of a resident, and occlusion problem is mitigated by using mirror. We use the reflection intensity and position information obtained from the target surface. Although it is not possible to identify all objects by additionally using reflection values, it would be easier to identify unknown objects if we can eliminate easily identifiable objects by reflectance. In addition, we propose a method for measuring the robot's pose using the tag which has the encoded reflection pattern optically identified by the LRF. Our experimental results validate the effectiveness of the proposed method.

Estimation of brain internal structures by deforming brain atlas using finite element method.

This paper presents a method for estimating the internal structures of a patient brain by deforming a standard brain atlas. Conventional deformation methods need several landmarks from the brain surface contour to fit the atlas to the patient brain shape. However, since the number and shapes of small sulci on the brain surface are different from each other, the determination of the accurate correspondence between small sulcus is difficult for experienced neurosurgeons. Moreover, the relationship between the surface shape and internal structure of the brain is unclear. Therefore, even if the deformed atlas is fitted to the patient brain shape exactly, the use of the deformed atlas does not always guarantee the reliable estimation of the internal structure of the patient brain. To solve these problems, we propose a new method for estimate the internal structure of a patient brain by the finite element method (FEM). In the deformation, our method select the landmarks from the contours of both the brain surface and the detectable internal structures from MR images.

Navigation system with real-time finite element analysis for minimally invasive surgery.

This paper presents a navigation system for minimally invasive surgery, especially laparoscopic surgery in which operates in abdomen. Conventional navigation systems show virtual images by superimposing models of target tissues on real endoscopic images. Since soft tissues within the abdomen are deformed during the surgery, the navigation system needs to provide surgeons reliable information by deforming the models according to their biomechanical behavior. However, conventional navigation systems don't consider the tissue deformation during the surgery. We have been developing a new real-time FEM-based simulation for deforming a soft tissue model by using neural network[1]. The network is called the neuroFEM. The incorporation of the neuroFEM into the navigation leads to improve the accuracy of the navigation system. In this paper, we propose a new navigation system with a framework of the neuroFEM.

Gait-based person identification robust to changes in appearance.

The identification of a person from gait images is generally sensitive to appearance changes, such as variations of clothes and belongings. One possibility to deal with this problem is to collect possible subjects' appearance changes in a database. However, it is almost impossible to predict all appearance changes in advance. In this paper, we propose a novel method, which allows robustly identifying people in spite of changes in appearance, without using a database of predicted appearance changes. In the proposed method, firstly, the human body image is divided into multiple areas, and features for each area are extracted. Next, a matching weight for each area is estimated based on the similarity between the extracted features and those in the database for standard clothes. Finally, the subject is identified by weighted integration of similarities in all areas. Experiments using the gait database CASIA show the best correct classification rate compared with conventional methods experiments.

Categorization of indoor places using the Kinect sensor.

The categorization of places in indoor environments is an important capability for service robots working and interacting with humans. In this paper we present a method to categorize different areas in indoor environments using a mobile robot equipped with a Kinect camera. Our approach transforms depth and grey scale images taken at each place into histograms of local binary patterns (LBPs) whose dimensionality is further reduced following a uniform criterion. The histograms are then combined into a single feature vector which is categorized using a supervised method. In this work we compare the performance of support vector machines and random forests as supervised classifiers. Finally, we apply our technique to distinguish five different place categories: corridors, laboratories, offices, kitchens, and study rooms. Experimental results show that we can categorize these places with high accuracy using our approach.

Real-time nonlinear FEM with neural network for simulating soft organ model deformation.

This paper presents a new method for simulating the deformation of organ models by using a neural network. The proposed method is based on the idea proposed by Chen et al. that a deformed model can be estimated from the superposition of basic deformation modes. The neural network finds a relationship between external forces and the models deformed by the forces. The experimental results show that the trained network can achieve a real-time simulation while keeping the acceptable accuracy compared with the nonlinear FEM computation.