Artificial Intelligence in Healthcare and Education

Community-based biomedical context to unlock agentic systems

This paper introduces BioContextAI, an open-source initiative that provides standardized access to validated biomedical knowledge to improve the reliability of AI systems in biomedical research.

A systematic protocol to identify 'clinical controls' for pediatric neuroimaging research from clinically acquired brain MRIs

EYE-Llama, an in-domain large language model for ophthalmology

This paper introduces EYE-Llama, an ophthalmology-specific large language model trained on in-domain data, which outperforms baseline models in medical question-answering tasks.

AI-HOPE: An AI-Driven conversational agent for enhanced clinical and genomic data integration in precision medicine research

AI-HOPE is a conversational AI platform that integrates clinical and genomic data using natural language, enabling analysis without programming expertise and producing statistical insights.

Artificial Intelligence's Contribution to Biomedical Literature Search: Revolutionizing or Complicating?

This study evaluates ChatGPT's utility for biomedical literature search compared to traditional methods, finding limitations in accuracy and consistency but potential with user-support tools.

Assessing Large Language Model Performance Related to Aging in Genetic Conditions

This study evaluates how large language models handle age-related aspects of 282 genetic conditions by generating medical vignettes and dialogues, finding no age-based biases but some performance differences.

Facilitating User Interaction with the Tuberculosis Mutation Catalogue using AI Tools

This paper evaluates AI models for interacting with a tuberculosis mutation catalogue, finding Google Gemini 2.5 Pro performs best in accuracy and reliability.

Conversational Artificial Intelligence for Translational Precision Medicine: Integrating Social Determinants of Health, Genomics, and Clinical Data with AI-HOPE-PM

AI-HOPE-PM integrates genomic, clinical, and social determinants of health data using conversational AI to enable real-time, multi-dimensional cancer analysis and uncover disparities in outcomes.

Reg-GPTTM: A Conversational AI Model for Enhanced Decision-Making in Regenerative Medicine

This paper introduces Reg-GPT, a conversational AI model fine-tuned for regenerative medicine applications, showing strong performance in accuracy, fluency, and engagement.

Classifying Domains, Benchmarking GPT-4, A Portuguese Dataset for Medical AI Q&A

The paper introduces a Portuguese medical question dataset for benchmarking AI models and classifying medical domains, showing GPT-4 achieves 90% accuracy and a BERTimbau-based classifier reaches 94% accuracy.

Large language models are universal biomedical simulators

This paper introduces SimulateGPT, a text-based simulator using large language models to predict biomedical outcomes without domain knowledge or manual tuning.

Using GPT-4 to write a scientific review article: a pilot evaluation study

This study evaluates GPT-4's ability to generate text, tables, and diagrams for biomedical review papers and suggests improvements for scientific use.

MedImg: a database for public medical image integration

The paper introduces MedImg, a database integrating 1,622,956 medical images from 103 datasets across 14 modalities and 9 organs, to support deep learning research in medical image analysis.

A RAG Chatbot for Precision Medicine of Multiple Myeloma

This paper introduces a RAG-based chatbot framework for precision medicine in Multiple Myeloma, integrating genomic data and biomedical literature to provide personalized treatment recommendations.

Empowering Personalized Pharmacogenomics with Generative AI Solutions

This study evaluates an AI assistant using GPT-4 and RAG to interpret pharmacogenomic results, showing improved performance over ChatGPT 3.5 in specialized queries.

Performance of GPT-4 on the American College of Radiology In-Service Examination

This study evaluates GPT-4's performance on a radiology exam, finding it performs between PGY-2 and PGY-3 levels but struggles with image-based questions and shows answer variability over time.

Performance of Large Language Models on Pharmacy Exam: A Comparative Assessment Using the NAPLEX

This study evaluates six LLMs on a pharmacy exam, finding GPT-4 performs best with 87.1% accuracy, suggesting potential impacts on pharmacy education and practice.

Genetic Discovery Enabled by A Large Language Model

This study uses a large language model to analyze gene-phenotype relationships in mouse genetic models, identifying causative genetic factors and generating novel hypotheses for spontaneous hearing loss.

Evaluating ChatGPT as an Agent for Providing Genetic Education

This study evaluates ChatGPT's ability to answer questions about genetic disorders, finding it provides accurate information but has limitations in understanding inheritance patterns.

Bioinformatics Illustrations Decoded by ChatGPT: The Good, The Bad, and The Ugly

This study evaluates ChatGPT's ability to interpret bioinformatics illustrations, finding it effective for explaining plots and applying biological knowledge but lacking in quantitative analysis.

Deep learning based phenotyping of medical images improves power for gene discovery of complex disease

This paper uses deep learning to phenotype knee osteoarthritis from medical images, improving GWAS power and polygenic risk score prediction compared to traditional methods.

Using artificial intelligence on dermatology conditions in Uganda: A case for diversity in training data sets for machine learning.

This paper evaluates an AI dermatology algorithm on Ugandan skin images, finding low diagnostic accuracy and emphasizing the need for diverse training data.

Analysis of large-language model versus human performance for genetics questions

This study compares the performance of a large-language model (ChatGPT) and humans on genetics questions, finding no significant difference overall but noting better memorization performance and inconsistent answers from the model.

Turing Test Inspired Method for Analysis of Biases Prevalent in Artificial Intelligence-Based Medical Imaging

The paper introduces a modified Turing Test to evaluate AI-based medical imaging systems for diagnostic accuracy and bias detection.

Description and clinical validation of a real-time AI diagnostic companion for fetal ultrasound examination

This paper describes a real-time AI diagnostic assistant for fetal ultrasound, SONIO, and validates its performance using postmortem phenotypes, achieving over 96% concordance with top-7 diagnoses.

Proof-of-principle neural network models for classification, attribution, creation, style-mixing, and morphing of image data for genetic conditions

The paper trains neural networks to classify, generate, and analyze images of rare genetic conditions, showing improved accuracy over experts and demonstrating style-mixing and morphing techniques.

Design and implementation of an intelligent framework for supporting evidence-based treatment recommendations in precision oncology

This paper designs an intelligent framework for evidence-based treatment recommendations in precision oncology, highlighting a six-step decision process and the importance of inference tracking.

An artificial intelligence-based first-line defence against COVID-19: digitally screening citizens for risks via a chatbot

Symptoma, a digital health assistant, accurately identifies COVID-19 cases with high precision using symptom input and outperforms simple questionnaires.

Efficient immune cell genome engineering with improved CRISPR editing tools

Doxycycline-inducible dual-color Cas9, purified lentivirus, puromycin-selected nickase HDR, integrase-deficient lentiviral HDR delivery and an all-in-one CRISPRa vector enable editing of macrophage and Jurkat T cells.

Breaking Medical Data Sharing Boundaries by EmployingArtificial Radiographs

This paper introduces a method to generate high-resolution artificial radiographs using generative models to overcome data sharing barriers in medicine.