Tomohiro Shibata

Krati Saxena, Tomohiro Shibata

Recommending safe and effective medication combinations from electronic health records (EHRs) is a core clinical AI problem, yet it remains difficult because patient trajectories are long, noisy, and clinically heterogeneous. Existing methods typically excel at either temporal modeling across visits or pharmacological knowledge integration (e.g., drug-drug interactions, DDIs), but rarely achieve both while robustly suppressing noise. We present GraphDiffMed, a knowledge-constrained medication recommendation framework built on dual-scale Differential Attention v2. Differential attention is applied at both intra-visit and inter-visit levels to filter spurious signals within encounters and across longitudinal history, while pharmacological constraints are incorporated during learning. Experiments on MIMIC-III and ablation studies show that this design consistently improves recommendation quality and ranking over strong baselines while achieving a more favorable safety performance balance. We further find that the strongest-performing configuration uses only demographic auxiliary features under our experimental setting. Overall, GraphDiffMed demonstrates that combining noise-aware attention with pharmacological constraints yields more reliable and clinically meaningful medication recommendation. We open-source our code at https://github.com/saxenakrati09/GraphDiffMed.

Septian Enggar Sukmana, Sang Won Bae, Tomohiro Shibata

Freezing of Gait (FOG) is a debilitating motor symptom commonly experienced by individuals with Parkinson's Disease (PD) which often leads to falls and reduced mobility. Timely and accurate prediction of FOG episodes is essential for enabling proactive interventions through assistive technologies. This study presents a reinforcement learning-based framework designed to identify optimal pre-FOG onset points, thereby extending the prediction horizon for anticipatory cueing systems. The model implements a Double Deep Q-Network (DDQN) architecture enhanced with Prioritized Experience Replay (PER) allowing the agent to focus learning on high-impact experiences and refine its policy. Trained over 9000 episodes with a reward shaping strategy that promotes cautious decision-making, the agent demonstrated robust performance in both subject-dependent and subject-independent evaluations. The model achieved a prediction horizon of up to 8.72 seconds prior to FOG onset in subject-independent scenarios and 7.89 seconds in subject-dependent settings. These results highlight the model's potential for integration into wearable assistive devices, offering timely and personalized interventions to mitigate FOG in PD patients.

Abdul Ali, Rikuo Kawamoto, Tomohiro Shibata

Parkinson's disease (PD) is a common neurodegenerative disease that affects motor and non-motor symptoms. Postural instability and freezing of gait (FOG) are considered motor symptoms of PD resulting in falling. In this study, we investigated the effect of simultaneous use of a robotic walker and a pneumatic walking assist device (PWAD) for PD patients on gait features. The pneumatic actuated artificial muscle on the leg and actuators on the walker produce mutual induced stimulation, allowing the user to suppress FOG and maintain a stable gait pattern while walking. The performance of the proposed system was evaluated by conducting an 8 [m] straight-line walking task by a healthy subject with (a) RW (robotic walker), (b) simultaneous use of an RW and a PWAD, and some gait features for each condition were analyzed. The increasing stride length and decreasing stance phase duration in the gait cycle suggest that simultaneous use of a robotic walker and a pneumatic walking assist device would effectively decrease FOG and maintain a stable gait pattern for PD patients.

John Noel Victorino, Naoto Fukunaga, Tomohiro Shibata

Coimagination method is a novel approach to support interactive communication for activating three (3) cognitive functions: episodic memory, division of attention, and planning. These cognitive functions are known to decline at an early stage of mild cognitive impairment (MCI). In previous studies about the coimagination method, experimenters tested different settings in different care institutions. Out of these experiments, various measures were introduced, analyzed, and presented. However, ease of changing configuration based on participants, and a quick assessment of captured data remained challenging. Also, several observers and measurers are needed to conduct the coimagination method. In this paper, we propose the initial design and development of an automated coimagination support system that can handle such challenges. We aim to have an automated coimagination support system that can be used easily either by healthy participants or elderly participants via a natural voice interface. In this paper, our focus is to measure how well our proposed features work with elderly participants. Preliminary experiments were conducted with healthy participants, and notably, with actual elder participants. Healthy participants experienced longer speaking round and question-and-answer round than with elderly participants; while, the latter had preparation time before the speaking round. In these preliminary experiments, our initial system showed the capability to handle different configurations. Healthy participants have operated the system using voice, while elderly participants managed to use the system with minimal assistance.