Medicine & Health

This study prompted ChatGPT to generate zero-shot appeal letters for cochlear implant insurance denials in asymmetric hearing loss or single-sided deafness across 30 prompt variations, had three cochlear implant providers score them against American Cochlear Implant Alliance guidelines, and checked citation accuracy; 96.6% of responses listed multiple benefits of cochlear implants and 79.3% partially aligned with the guidelines, but 51.7% contained hallucinated benefits and only six of 96 citations (6.3%) accurately referenced peer-reviewed sources, with the rest hallucinated (57.3%), erroneous (24%), or irrelevant (10%), indicating that human verification is needed before clinical or advocacy use.

This article reviews advances in applying artificial intelligence across the full chain of head and neck tumor care from screening and diagnosis to follow-up management, noting that endoscopic images, CT/MRI imaging, digital pathology, ultrasound, and multi-omics data are the main data modalities, that AI-assisted endoscopic interpretation and imaging analysis can improve lesion recognition accuracy, reduce operator dependence, and support non-invasive prediction of occult lymph node metastasis, while current evidence comes mostly from single-center retrospective studies and prospective evidence for treatment response prediction remains limited.

This work proposes a novel architecture that integrates a local feature extraction module with a global context model, in which the Mel_Grouper module serves as a front-end to enhance local pathological representations and its output is fed into the Transformer-based Mel_Encoder to fuse global context, achieving improvements of 3.37%, 3.06%, 6.83% and 5.36% over the previous best results on the four subtasks of the SJTU Paediatric Respiratory Sound (SPRSound) dataset, with further experiments on a real-world paediatric respiratory sound dataset.

In routine use of a surgical AI platform, the study automatically assigned Parkland Grading Scale scores across 249 consecutive laparoscopic cholecystectomies, grouped them into Low (PGS 1-2, n=78) and High (PGS 3-5, n=171) severity, compared surgical outcomes, and evaluated model F1, discrimination (AUC 0.932 and 0.896) and calibration against two surgeons' independent double reviews of a 75-video sample, finding that the High group had older age, higher ASA scores, more cholecystitis and urgent surgery, longer operative duration, more intraoperative events and bailouts, longer hospital stay, and more 90-day major complications and readmissions, with operative duration and hemorrhage-related events remaining significant after case-control matching (n=84).

In a single-center retrospective study, an attention-based multi-task deep learning model combining a ResNet+CBAM imaging branch with a DNN+Transformer clinical-feature branch used contrast-enhanced neck CT to jointly predict lymph node malignancy and primary site, achieving in the test set (n=457) an AUC of 0.851 for benign/malignant prediction, a Micro-AUC of 0.819 for primary-site prediction, and Top-1/Top-3 accuracy of 79%/93%, and in a real-world CMSCCUP cohort (172 cases) a Micro-AUC of 0.809 and Top-3 accuracy of 88%, significantly improving junior physicians' predictive accuracy (all P<0.001).

This Commentary reviews recently published sleep foundation models by comparing their training cohorts, assessment frameworks, and reported performance, and applies an existing sleep foundation model without fine-tuning to an independent cohort of patients with Narcolepsy Type 1 (n = 51) and healthy controls (n = 28), finding modest zero-shot sleep-staging performance below supervised methods on the same cohort and minimal improvement in disorder classification from PSG-derived embeddings beyond demographic baselines, concluding that sleep foundation models are not yet suitable for clinical deployment.

This article argues that prompt injection is a failure mode distinct from accuracy, bias, and hallucination—a model performing exactly as instructed by an instruction the clinician neither wrote nor can see—arising from a fundamental property of current language-model architectures that receive an undifferentiated stream of tokens and possess no mechanism for distinguishing content that carries authority from content that does not, with medicine particularly exposed because the clinical record is assembled from material originating outside the institution, including referral correspondence, patient-entered messages, external reports, scanned documents, and imaging acquired elsewhere; the authors argue that prompt injection warrants classification as a patient safety hazard with an articula

Using data from the TRANQUIL multicenter randomized controlled trial, this study estimated minimal clinically important differences (MCIDs) for the modified activities of daily living (mADL) and mADL11 scales of TETRAS in essential tremor via anchor-based approaches (CGI-I and PGI-I) and ROC analyses, recommending patient-report-based MCIDs of -6.2 for mADL and -4.1 for mADL11.

Using ADNI data, this study builds a patient-level multimodal fusion framework based on ResNet50 transfer learning that concatenates MRI and PET imaging features with clinical variables (ADAS11, ADAS13, APOE4, age, sex, education, MMSE total) in a fully connected network to produce three-way AD/MCI/CN predictions, with Grad-CAM heatmaps for explanation; the ablation shows accuracy rising as modalities are added, from 39.61% for MRI alone and 51.72% for PET alone to 66.67% for clinical-only, 54.17% for MRI+PET, and 79.17% for the full three-modality model on a 24-patient test cohort (95% Wilson interval roughly 59.5%-90.8%), with the addition of clinical data producing the single largest gain.