Medicine & Health

469 items

Journal of Medical Internet Research Sep 14, 2026
This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
Anonymization of Portuguese Clinical Notes Using Large Language Models and Quantum-Enhanced Hybrid Architectures: Comparative Evaluation Study
This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
Journal of Medical Internet Research Sep 14, 2026
Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
How Generative AI Video Models Depict Depression: A Mixed Methods Study of OpenAI's Sora 2
Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
Journal of Chemical Information and Modeling Sep 14, 2026
Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Is AI Capable of Real-World Drug Discovery?
Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Acta Pharmacologica Sinica Sep 14, 2026
This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
Therapeutic cancer vaccines: development, challenges, and future perspectives
This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
Physics in Medicine and Biology Sep 14, 2026
In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
GPT-assisted radiomic modeling for predicting pathological complete response to neoadjuvant chemoimmunotherapy in head and neck squamous cell carcinoma
In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
medRxiv Sep 14, 2026
This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
Ischemic Stroke Detection, Segmentation, and Volume Estimation from Multi-sequence MRI with Missing Sequences
This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
JMIR Formative Research Sep 14, 2026
This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
Adaptive Optics Image Analysis Using Generative AI (GPT-4) Scripting: An Exploratory Study
This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
Academic Radiology Sep 14, 2026
This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
Who Read It First? Documenting Independent Judgment in AI-Assisted Radiology
This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
Journal of Medical Internet Research Sep 14, 2026
Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
What Single-Topic Summaries Miss in Hospital Reviews: Aspect-Level Evaluative Structure Using Generative Pretrained Transformer-Based Sentiment Analysis
Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
New Biotechnology Sep 14, 2026
This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8
This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8
Sequence optimization targeting mRNA stability enhances monoclonal antibody titers in CHO cells
This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8
This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8

Page 44 · showing 10