Digital twins in medicine offer personalized models that replicate patient-specific anatomy and physiology, enabling accurate diagnosis, individualized treatment planning, and predictive simulations. However, constructing reliable digital twins, particularly for complex anatomical structures like the human cochlea, requires robust generative modeling methods capable of synthesizing high-resolution medical images. In this paper, we propose a novel Langevin Variational Autoencoder (LVAE) based generative framework, integrating stochastic flows to enhance the expressiveness and computational efficiency of variational inference. By employing a quasi-symplectic Langevin flow, the proposed LVAE overcomes the limitations of existing approaches, reducing computational overhead associated with Hessian computations. Experimental evaluations on a large-scale cochlea CT dataset demonstrate superior performance of the LVAE in terms of inference accuracy, convergence speed, and image realism compared to existing stochastic flow-based VAEs. Our approach provides an effective foundation for building clinically reliable cochlear digital twins, thereby advancing personalized medicine and improving clinical outcomes for patients with hearing impairments.
Digital Twin (DT) technology offers real-time monitoring, simulation, optimization, and precise forecasting. However, its theoretical framework and practical implementation, particularly in digital random controlled trials (RCT), remain underdeveloped, limiting its full potential. Therefore, we seek to develop an application of digital twins with the virtual reality using data from four longitudinal RCT in Massachusetts. As case studies, we create digital twins of two individuals from these RCT. This application of digital twins provides an avenue for a more personalized healthcare experience for patients, the enhancement of medical simulations, and a visualization for predictive analytics. In the future, we will move from virtual reality to extended reality with AI-generated digital twin models.
Fundus imaging in preterm infants is critical for diagnosing retinopathy of prematurity (ROP) but requires capturing multiple fields-of-view (FOVs), imposing significant distress on infants and operational burdens on clinicians. To address this, we propose a Topology-Aware Conditional Latent Diffusion Model (TA-CLDM) that synthesizes patient-specific, multi-view fundus images from a single input, conditioned on textual FOV prompts (e.g., "left eye, nasal view"). Our framework integrates a latent diffusion model (LDM) with a topology-aware loss that leverages persistent homology to preserve vascular continuity-ensuring synthesized vessels retain diagnostically critical structures like branching patterns and connectivity. Trained on a dataset of 2,862 preterm infant fundus images (954 patients, 3 FOVs/patient), TA-CLDM achieves a Frechet Inception Distance (FID) of 47.93 outperforming baseline LDMs by 19.91 FID. By reducing the need for repeated imaging, TA-CLDM alleviates infant distress and institutional costs while maintaining diagnostic accuracy. This work pioneers the integration of text-guided synthesis and topological preservation in medical imaging, offering a scalable solution for neonatal care.
Freezing of Gait (FoG) is a debilitating symptom of Parkinson's disease (PD), often triggered by specific environmental factors. Existing studies focus solely on detecting FoG after it occurs. This reactive approach fails to prevent episodes or mitigate associated risks ahead of time. To address this limitation, we propose TriggerFinder, a novel real-time FoG trigger detection system powered by a multimodal large language model (MLLM). To adapt MLLM for FoG trigger detection, we instruction-tune an open-source MLLM using a newly curated dataset comprising five common FoG triggers identified by domain experts: floor pattern changes, tight turns, dual-tasking, narrow passages, and distractions. For ensuring realtime detection, we offload the MLLM inference tasks to an edge server. We evaluate Trigger-Finder on a new trigger benchmark dataset collected from open environments. The evaluation results show that our approach significantly outperforms the baseline methods in both accuracy and inference speed. It identifies FoG triggers in approximately 0.8 seconds, enabling real-time detection in open environments. Additionally, two case studies illustrate our method enhances both accuracy and interpretability, demonstrating its potential to prevent or mitigate FoG in advance.
Mental health status has been identified as a key factor leading to productivity deficits in individuals. Several types of research in different platforms, including the All of Us workbench, are ongoing navigating from the diagnosis of mental health disorders to the treatment of such problems. Nevertheless, there is a major gap in the current literature with regard to the investigation of the underlying problems of the underrepresented groups of people, the predominant focus being on immediate solutions. To shed light on this issue, we performed an analysis of the critical features of mental health disorders among different populations, such as White and Black or African American populations, using a variety of feature selection strategies. Furthermore, we examined the disparities informed by these essential characteristics to uncover the root causes. Moreover, we investigated the multicollinearity of features among the underrepresented subpopulations and offered a comprehensive analysis from their perspective.
