JMIR Medical Informatics最新文献

We used the free artificial intelligence (AI) tool Google NotebookLM, powered by the large language model Gemini 2.0, to construct a medical decision-making aid for diagnosing and managing airway diseases and subsequently evaluated its functionality and performance in a clinical workflow. After feeding this tool with relevant published clinical guidelines for these diseases, we evaluated the feasibility of the system regarding its behavior, ability, and potential, and we created simulated cases and used the system to solve associated medical problems. The test and simulation questions were designed by a pulmonologist, and the appropriateness (focusing on accuracy and completeness) of AI responses was judged by 3 pulmonologists independently. The system was then deployed in an emergency department setting, where it was tested by medical staff (n=20) to assess how it affected the process of clinical consultation. Test opinions were collected through a questionnaire. Most (56/84, 67%) of the specialists' ratings regarding AI responses were above average. The interrater reliability was moderate for accuracy (intraclass correlation coefficient=0.612; P<.001) and good on completeness (intraclass correlation coefficient=0.773; P<.001). When deployed in an emergency department (ED) setting, this system could respond with reasonable answers, enhance the literacy of personnel about these diseases. The potential to save the time spent in consultation did not reach statistical significance (Kolmogorov-Smirnov [K-S] D=0.223, P=.24) across all participants, but it indicated a favorable outcome when we analyzed only physicians' responses. We concluded that this system is customizable, cost efficient, and accessible to clinicians and allied health care professionals without any computer coding experience in treating airway diseases. It provides convincing guideline-based recommendations, increases the staff's medical literacy, and potentially saves physicians' time spent on consultation. This system warrants further evaluation in other medical disciplines and health care environments.

Background: A rapidly aging population led to an increase in the number of patients with chronic diseases and polypharmacy. Although investigations on the appropriate number of drugs for older patients have been conducted, there is a shortage of studies on polypharmacy criteria in older inpatients from multiple institutions.

Objective: The aim of this study was to examine the patterns of polypharmacy and determine the criteria for the number of drugs defining polypharmacy in the geriatric inpatient population.

Methods: Electronic health records of 4 medical institutions for patients aged 65 years and older hospitalized between January 1, 2012, and December 31, 2020, were analyzed for the study. The maximum number of drugs prescribed was obtained for each patient and, along with a literature review, was used to determine the appropriate polypharmacy level for our population.

Results: We suggest a 4-level polypharmacy category system consisting of nonpolypharmacy, polypharmacy, major polypharmacy, and excessive polypharmacy based on a review of international guidelines and polypharmacy literature. Application of this system to our study population showed that the major polypharmacy category (use of 10-19 concurrent drugs) was an appropriate threshold for polypharmacy in hospitalized patients versus the traditional threshold of 5 or more concurrent drugs. The tendency of our study population to have a higher disease and drug count supports this threshold. Frequently prescribed therapeutic subgroups in this category were antibacterials for systemic use, anesthetics, and cardiac therapy.

Conclusions: This study proposes a polypharmacy categorization system for older inpatients, which differs from the common definition of the concomitant prescription of 5 or more drugs. The older population tends to have severe conditions including those requiring major surgeries; therefore, a drug count corresponding to the definition of major polypharmacy is appropriate.

Background: Early life health risks can shape long-term morbidity trajectories, yet prevailing pediatric risk assessment paradigms are often fragmented and insufficiently capable of integrating heterogeneous data streams into actionable, individualized profiles.

Objective: This study aimed to design, implement, and validate an artificial intelligence-driven framework that fuses multimodal pediatric data and leverages advanced natural language processing and ensemble learning to improve early, accurate stratification of key pediatric health risks.

Methods: A retrospective dataset of over 40,000 pediatric participants aged 2-8 years was used to train and evaluate the framework. Data were split into training, validation, and test sets (70%, 15%, and 15%, respectively) with a temporally mindful partitioning strategy to approximate prospective evaluation. Baseline comparators included traditional statistical and machine learning models, and the statistical significance of area under the receiver operating characteristic curve (AUC-ROC) differences was assessed using the DeLong test.

Results: The proposed Bidirectional Encoder Representations From Transformers-based model achieved an AUC-ROC of 0.85 (95% CI 0.82-0.88), sensitivity of 0.78, specificity of 0.80, and F1-score of 0.75 on the test set, outperforming multiple baseline models. In an additional manual comparison evaluation, automated and expert assessments aligned with 78% accuracy (78/100), and most discrepancies arose in "equivalent" cases.

Conclusions: This study provides a validated, artificial intelligence-driven, multimodal pediatric health risk stratification framework that translates heterogeneous child health data into clinically actionable risk profiles, demonstrating strong discriminative performance and meaningful agreement with expert assessment. The framework supports proactive, individualized pediatric care and offers a scalable foundation for further validation across broader populations and longitudinal follow-up.

Background: Given the highly heterogeneous biology of breast cancer, a more effective noninvasive diagnostic tool that unravels microscopic histopathology patterns is urgently needed.

Objective: This study aims to identify cancerous regions in ultrasound images of breast cancer via convolutional neural network based on registered grayscale ultrasound images and readily accessible biopsy whole slide images (WSIs).

Methods: This single-center study prospectively included participants undergoing ultrasound-guided core needle biopsy procedures for Breast Imaging Reporting and Data System category 4 or 5 breast lesions for whom breast cancer was pathologically confirmed from July 2022 to February 2023 consecutively. The basic information, ultrasound image data, biopsy tissue specimens, and corresponding WSIs were collected. After core needle biopsy procedures, the stained breast tissue specimens were sliced and coregistered with an ultrasound image of a needle tract. Convolutional neural network models for identifying breast cancer cells in ultrasound images were developed using FCN-101 and DeepLabV3 networks. The image-level predictive performance was evaluated and compared quantitatively by pixel accuracy, Dice similarity coefficient, and recall. Pixel-level classification was illustrated through confusion matrices. The cancerous region in the testing dataset was further visualized in ultrasound images. Potential clinical applications were qualitatively assessed by comparing the automatic segmentation results and the actual pathological tissue distributions.

Results: A total of 105 participants with 386 ultrasound images of breast cancer were included, with 270 (70%), 78 (20.2%), and 38 (9.8%) images in the training, validation, and test datasets, respectively. Both models performed well in predicting the cancerous regions in the biopsy area, whereas the FCN-101 model was superior to the DeepLabV3 model in terms of pixel accuracy (86.91% vs 69.55%; P=.002) and Dice similarity coefficient (77.47% vs 69.90%; P<.001). The two models yielded recall values of 54.64% and 58.46%, with no significant difference between them (P=.80). Furthermore, the FCN-101 model had an advantage in predicting cancerous regions, while the DeepLabV3 model achieved more accurate predictive pixels in normal tissue (both P<.05). Visualization of cancerous regions on grayscale ultrasound images demonstrated high consistency with those identified on WSIs.

Conclusions: The technique for spatial registration of breast WSIs and ultrasound images of a needle tract was established. Breast cancer regions were accurately identified and localized on a pixel level in high-frequency ultrasound images via an advanced convolutional neural network with histopathologic WSI as the reference standard.

Background: Deep learning models have shown strong potential for automated fracture detection in medical images. However, their robustness under varying image quality remains uncertain, particularly for small and subtle fractures, such as scaphoid fractures. Understanding how different types of image perturbations affect model performance is crucial for ensuring reliable deployment in clinical practice.

Objective: This study aimed to evaluate the robustness of a deep learning model trained to detect scaphoid fractures in radiographs when exposed to various image perturbations. We sought to identify which perturbations most strongly impact performance and to explore strategies to mitigate performance degradation.

Methods: Radiographic datasets were systematically modified by applying Gaussian noise, blurring, JPEG compression, contrast-limited adaptive histogram equalization, resizing, and geometric offsets. Model accuracy was evaluated across different perturbation types and levels. Image quality was quantified using peak signal-to-noise ratio and structural similarity index measure to assess correlations between degradation and model performance.

Results: Model accuracy declined with increasing perturbation severity, but the extent varied across perturbation types. Gaussian blur caused the most substantial performance drop, whereas contrast-limited adaptive histogram equalization increased the false-negative rate. The model demonstrated higher resilience to color perturbations than to grayscale degradations. A strong linear correlation was found between peak signal-to-noise ratio-structural similarity index measure and accuracy, suggesting that better image quality led to improved detection. Geometric offsets and pixel value rescaling had minimal influence, whereas resolution was the dominant factor affecting performance.

Conclusions: The findings indicate that image quality, especially resolution and blurring, substantially influences the robustness of deep learning-based fracture detection models. Ensuring adequate image resolution and quality control can enhance diagnostic reliability. These results provide valuable insights for designing more accurate and resilient medical imaging models under real-world variability.

Background: When used correctly, electronic medical records (EMRs) can support clinical decision-making, provide information for research, facilitate coordination of care, reduce medical errors, and generate patient health summaries. Studies have reported large differences in the quality of EMR data.

Objective: Our study aimed to develop an evidence-based set of electronically extractable quality indicators (QIs) approved by expert consensus to assess the good use of EMRs by general practitioners (GPs) from a medical perspective.

Methods: The RAND-modified Delphi method was used in this study. The TRIP and MEDLINE databases were searched, and a selection of recommendations was filtered using the specific, measurable, assignable, realistic, and time-bound principles. The panel comprised 12 GPs and 6 EMR developers. The selected recommendations were transformed into QIs as percentages.

Results: A combined list of 20 indicators and 30 recommendations was created from 9 guidelines and 4 review articles. After the consensus round, 20 (100%) indicators and 20 (67%) recommendations were approved by the panel. All 20 recommendations were transformed into QIs. Most (16, 40%) QIs evaluated the completeness and adequacy of the problem list.

Conclusions: This study provided a set of 40 EMR-extractable QIs for the correct use of EMRs in primary care. These QIs can be used to map the completeness of EMRs by setting up an audit and feedback system, and to develop specific (computer-based) training for GPs.