In this study, the development of a deep learning approach for distinguishing cardiomegaly in chest X-ray images and its validation process are presented. Typically, radiologists diagnose cardiomegaly by examining X-ray images. However, their interpretations can vary owing to subjective judgments, and mild cardiomegaly can be missed. For this reason, there is ongoing research into the use of AI-based deep learning algorithms as an adjunct to X-ray interpretation. In this study, radiologists collected 10,000 public images, from which 718 useful images were selected to create a dataset. A DenseNet121 algorithm was then used to develop an AI model for cardiomegaly detection. The results demonstrate an accuracy of 0.95, a recall of 0.91, and an F1 score of 0.94. Additional validation was performed to ensure the accuracy of the cardiomegaly detection model. The validation methods included saliency maps and guided backpropagation, which indicated the significance of the model. In conclusion, this study demonstrates the potential to develop a high-quality AI model with fewer data than previous studies, suggesting its applicability in the medical field.

Sleep apnea has emerged as a significant health issue in modern society, with self-diagnosis and effective management becoming increasingly important. Among the most renowned methods for self-diagnosis, the STOP-BANG questionnaire is widely recognized as a simple yet effective tool for diagnosing and assessing the risk of sleep apnea. However, its sensitivity and specificity have limitations, necessitating the need for tools with higher performance. Consequently, this study aimed to enhance the accuracy of sleep apnea diagnoses by integrating machine learning with the STOP-BANG questionnaire. Research through actual cases was conducted based on the data of 262 patients undergoing polysomnography, confirming sleep apnea with a STOP-BANG score of ≥3 and an Apnea–Hypopnea Index (AHI) of ≥5. The accuracy, sensitivity, and specificity were derived by comparing Apnea–Hypopnea Index scores with STOP-BANG scores. When applying machine learning models, four hyperparameter-tuned models were utilized: K-Nearest Neighbor (K-NN), Logistic Regression, Random Forest, and Support Vector Machine (SVM). Among them, the K-NN model with a K value of 11 demonstrated superior performance, achieving a sensitivity of 0.94, specificity of 0.85, and overall accuracy of 0.92. These results highlight the potential of combining traditional STOP-BANG diagnostic tools with machine learning technology, offering new directions for future research in self-diagnosis and the preliminary diagnosis of sleep-related disorders in clinical settings.

Background/Objectives: This study demonstrates improved generalization in pressure-ulcer stage classification. In medical imaging, training data are often scarce and disease specific. For skin conditions such as pressure ulcers, variation in camera to subject distance, resolution, illumination, and viewpoint across photographers reduces accuracy in clinical use. Methods: We developed a YOLOv7-based pressure ulcer stage classification model by employing a two-phase training strategy. Phase 1 was trained on the full dataset stratified by pressure-ulcer stage. Phase 2 was trained in saliency-guided images augmented with clinically plausible noise, including healing areas and white keratin. The added dataset comprised 296 images obtained by randomly sampling 30% from stages 1 through 3 of the full dataset. Results: The accuracy of the 38 newly acquired hospital images increased from 75% in Phase 1 to 89% in Phase 2. Five-fold cross-validation demonstrated stable performance (mAP@0.5: 86.20% ± 2.28%), confirming reproducibility. This exceeds by more than five percentage points the performance reported for pressure-ulcer staging models in prior studies conducted in clinical deployment settings. Conclusions: These findings suggest that curriculum learning combined with noise-enriched augmentation can improve generalization in clinical environments. Our results demonstrate that clinically informed data augmentation is a key factor in enhancing the model’s clinical generalization. Accordingly, the proposed approach provides a practical path to enhancing clinical usability in data-limited medical imaging.

Background/Objectives: Pressure injury stage classification was performed using object detection models to address inconsistencies in clinical assessment due to variability in nurses’ experience and education levels. Methods: A dataset of 1282 pressure injury images from a medical institution was used to train and compare five representative architectures, YOLOv5x, YOLOv7, YOLOv8x, YOLOv8n, and YOLOv11x, and Faster R-CNN across Stages 1–4, excluding Deep Tissue Injury and unclassified cases. A mobile application incorporating YOLOv7 was deployed at Eulji University Daejeon Medical Center and tested by 10 nurses over 2 weeks, processing 46 cases. Results: YOLOv7 demonstrated superior performance with mAP@0.5 of 0.97 and mAP@0.5:0.95 of 0.68, achieving 93% accuracy for Stage 2 classification, the most challenging diagnostic category. Clinical validation demonstrated 87% diagnostic accuracy, 4.0/5 user satisfaction, and workflow improvement with assessment time reduced from 4–6 min to 1 min. The application proved valuable as both a diagnostic support tool and educational resource for novice nurses, with zero critical misclassifications recorded. Conclusions: This study establishes the practical utility of AI-based pressure injury classification systems in clinical practice and their potential for enhancing nursing competency and workflow efficiency.

Background/Objectives: A four-class stage classifier can assign a pressure ulcer stage to an erroneous detection. We evaluated supervised candidate verification and its rejection costs. Methods: This retrospective challenge-set study used a frozen detector and stage classifier with an existing binary gate threshold of 0.70. Negative evaluation data comprised 1138 Normal-labeled skin images from 347 source photographs and 328 non-pressure-wound images. A stronger source-family audit reduced the original 112-image positive subset to 94 images from 93 families; these still overlapped detector development. Additional experiments compared three source-grouped binary classifiers and paired detector fine-tuning with or without mined negative images. Results: The historical gate reduced skin images with false-positive outputs from 269/1138 (23.6%) to 54/1138 (4.7%). In the corrected positive subset, 106/107 detected lesions were retained; overall lesion recall changed from 107/110 to 106/110. Both historical false-negative lesions remain in the safety review. Newly trained You Only Look Once (YOLO) 26s, ResNet18, and MobileNetV3-small models achieved internal test-positive retention of 97.4–98.5% after separate calibration. Adding mined negatives directly to detector fine-tuning also reduced false-positive outputs. The historical gate accepted 122/147 matched non-pressure-wound candidates. Conclusions: Negative supervision reduced false-positive outputs through both modular verification and detector retraining. The findings do not establish superiority of a separate gate, patient-level generalization, clinical acceptability of missed lesions, or pressure ulcer etiology discrimination.