Ai In Medicine
MedImageEdu: A New Benchmark for Multi-Modal Patient Education
The MedImageEdu benchmark addresses limitations in existing medical multimodal tasks by focusing on multi-turn, evidence-grounded radiology patient education. It simulates doctor-patient interactions, incorporating visual support and evaluating agents across five dimensions. This benchmark highlight…
Latent Attention Masked Autoencoders for Multi-View Echocardiography Outperform Standard MAE in Cardiac Assessment
LAMAE introduces a novel foundation model architecture that leverages latent attention to integrate information across varying frames and views in echocardiography. This method enables the reconstruction of a holistic cardiac representation from incomplete observations, a significant improvement ove…
Wearable Data and Multi-Instance Learning for Frailty Estimation in Elderly Cancer Patients
Frailty in elderly cancer patients significantly impacts treatment outcomes, but current assessment methods are often insufficient. This study proposes and validates a multimodal wearable framework utilizing smartwatch and chest strap data to estimate frailty-related functional changes between clini…
Wearable-Based Stress Detection for Elderly Cancer Patients Shows Moderate Agreement with Self-Reported Scores
This study explores the use of multimodal wearable data (smartwatch and ECG sensor) to estimate perceived stress in elderly breast cancer patients. By transforming wearable streams into visual representations and employing an attention-based multiple instance learning (MIL) approach with a Tiny-BioM…
Orthogonal Subspace Decomposition for Multimodal MRI-PET Fusion Reveals Modality-Specific Information
This paper introduces a novel subspace decomposition framework for multimodal imaging analysis, specifically focusing on MRI and PSMA PET data. The method separates PET uptake into an MRI-explainable physiological component and an orthogonal residual, which represents information unique to PET. This…
Ensemble Deep Clustering Enhances EHR Analysis for Heart Failure
Traditional clustering methods often outperform deep learning approaches when analyzing tabular EHR data, as deep learning models are primarily designed for image processing. This paper introduces an ensemble-based deep clustering framework that aggregates multiple embedding dimensions, improving pe…
AI-Powered Personalized Medicine Accelerates Drug Discovery Beyond Traditional Pharmaceutical Pipelines
A single individual, leveraging AI tools like ChatGPT and AlphaFold with a modest investment, successfully designed a custom mRNA cancer vaccine for a pet, achieving significant tumor reduction. This case demonstrates AI's potential to democratize and accelerate drug discovery, outpacing conventiona…
STARC-9: A Diverse Dataset for Colorectal Cancer Histopathology Classification
STARC-9 is a new large-scale dataset for multi-class tissue classification in colorectal cancer (CRC) histopathology. It addresses limitations of existing datasets by providing morphologically diverse, high-quality image tiles across nine clinically relevant tissue classes. The dataset was construct…