Skin lesion images are relatively easy to obtain, but variations in imaging conditions, particularly lesion positioning, can cause inconsistencies in deep learning model activations, impacting diagnostic reliability. In this study, a robust deep-learning model for melanoma prediction was developed using class-agnostic activation maps (CAAMs) for enhanced diagnostic accuracy and reliability by addressing issues related to image variability and transformation robustness. The international skin imaging collaboration (ISIC) 2017 and 2019 datasets, focusing on melanoma and nevus, were used. The performance was evaluated using the area under the receiver operating characteristic curve (AUROC), while Dice scores were used for robustness evaluation. The proposed model achieved an AUROC of 0.954 for ConvNeXt on the ISIC 2019 dataset, with Dice scores of 0.664 and 0.457 for ConvNeXt and ResNet, respectively. For the ISIC 2017 dataset, the proposed model achieved an AUROC of 0.843 for ConvNeXt, with Dice scores of 0.557 and 0.306 for ConvNeXt and ResNet, respectively. The proposed CAAM-based method improved the melanoma prediction and lesion recognition accuracy by ensuring robust and consistent CAAMs.

Thyroid cancer treatment often involves thyroidectomy, which can lead to significant cosmetic issues due to scarring. This study proposes artificial intelligence to predict the severity and characteristics of post-thyroidectomy scars using a dataset of 24,731 images from 5,306 patients. The deep learning model, ResNet 50, was trained to assess scar severity with a mean absolute error of 1.036 and a Pearson correlation coefficient of 0.640. The study highlights the model's performance variations across different scar types and underscores the challenge of data imbalance in severity levels. The AI-driven approach aims to guide appropriate treatment strategies for post-thyroidectomy scars.

The growing number of endoscopic surgery has underscored the critical need for accuracy and reliability in surgical imaging systems. However, endoscopic procedures face significant challenges, including limited lighting and smoke generated during surgeries. These factors obstruct visibility and compromise the precision of surgical interventions. This study focuses on developing an artificial intelligence-based system to enhance endoscopic image quality and effectively address these challenges. The proposed system integrates image enhancement algorithms with a smoke removal model. To mitigate low-light conditions and emphasize vascular structures, image enhancement was performed using various filters. For smoke removal, a generative adversarial network-based model was employed. Test results demonstrated that the system outperformed existing smoke-specific removal models, achieving PSNR and SSIM scores of 26.52 and 0.9119, respectively. These findings highlight the potential of this system to significantly improve the reliability and efficiency of endoscopic surgeries.

Early diagnosis and treatment of pancreatic diseases are crucial for improving patient outcomes. Abdominal CT scans provide high-resolution images essential for accurate diagnosis, but analyzing these images is challenging due to the pancreas's complex anatomy and requires expert labor. We propose an anomaly detection algorithm using a latent diffusion model to generate disease-free CT images from abnormal slices. In this study, we used data from 579 patients, including 478 healthy individuals and 101 with pancreatic diseases, training the model on healthy slices only. About 98,000 2D images of 512x512 pixels were used with optimized window-level settings for contrast enhancement. The proposed algorithm achieved an accuracy of 0.837 and an AUC of 0.890. In additional experiments with random slices from 50 normal and 50 abnormal cases, it achieved an accuracy of 0.830 and an AUC of 0.878, demonstrating potential for rapid, effective diagnostic support in clinical settings.

This study introduces an automated wound tissue segmentation algorithm utilizing deep learning models, including UNet base models, trained on a dataset of wound image. Among the models, Attention UNet demonstrated superior performance with the highest sensitivity (0.622) and Dice score (0.608), enabling precise segmentation of wound tissues and providing a robust tool for clinical wound assessment and monitoring.

Recent advancements in deep learning for automatic teeth alignment adopt a method that extracts dental features through 3D scans and predicts teeth positions using quaternions. In this study, we converted 3D dental scan data into point clouds, then used these point clouds as input to extract features and align teeth through regression analysis that predicts quaternions. Various models were trained using front tooth data, and the results were compared with front tooth data that were not used for training. Experimental results confirmed successful alignment of the front teeth, visualized along the alignment axis. These advancements suggest a promising direction for more precise automatic teeth alignment using deep learning.

Vitiligo, a skin condition characterized by depigmented patches, significantly affects patients' quality of life. Current evaluation methods, such as the Facial Vitiligo Area Scoring Index (F-VASI), rely on subjective clinical assessments and often require specialized equipment. This study introduces a novel approach for the automated assessment of facial vitiligo using multi-view UV images and a landmark-based facial region subdivision algorithm. By leveraging detected facial landmarks, binary masks were created to segment the face into subregions, enabling detailed evaluation of depigmentation patterns without additional equipment. This work represents a significant step toward objective and efficient vitiligo evaluation.

Benign paroxysmal positional vertigo is a common vestibular disorder characterized by nystagmus, an involuntary eye movement indicative of vestibular dysfunction. Accurate diagnosis of nystagmus remains challenging due to its complex patterns and the dependency on examiner expertise. This study introduces a novel classification framework combining wavelet transform and deep learning to enhance the diagnostic accuracy of nystagmus in videonystagmography. Pupil trajectories were extracted from videonystagmography data and preprocessed to remove motion artifacts. Wavelet transform was applied to convert time-series signals into two-dimensional time-frequency representations. Convolutional neural network was utilized to classify nystagmus presence and direction, achieving an overall accuracy of 87% and sensitivity of 88% for directional classification. The integration of wavelet transform with convolutional neural networks effectively captured temporal and frequency features, demonstrating superior performance over conventional methods. This method holds potential to revolutionize nystagmus diagnosis, providing clinicians with a reliable and efficient tool for early detection of vestibular disorders.

Efficient patient flow management is essential for improving hospital operations, reducing delays, and enhancing the patient experience. This study presents a Transformer-based model designed to predict patient movement and treatment pathways using historical log data. By employing a hierarchical prediction approach, the model forecasts patients’ next steps across four levels(Section, Group, Producer, and Activity)with superior sequential accuracy. The model was trained on the BPIC 2011 dataset, which includes diagnostic and treatment records from a Dutch academic hospital. Data preprocessing involved sequence filtering, feature engineering, and a sliding window approach to create time-series input-output pairs. Proposed model demonstrated strong performance, outperforming baseline LSTM and standard Transformer models in F1 scores and sequential prediction accuracy. This preliminary study highlights the feasibility of using predictive models for individual patient movements. Future research aims to scale this approach to handle multiple patient pathways simultaneously, enabling real-time density forecasting and bottleneck management in high-demand areas, such as emergency rooms. These findings underscore the potential of predictive modeling to optimize hospital efficiency and patient care.

Variant Angina (VA) is a condition characterized by chest pain caused by atherosclerotic spasms and contractions. It can affect younger individuals without typical risk factors and is often accompanied by symptoms like sweating, tachycardia, and low blood pressure. In severe cases, VA can result in life-threatening complications such as myocardial infarction or arrhythmia, highlighting the importance of accurate diagnosis and early prevention. Although various diagnostic methods exist in clinical practice, diagnosing VA remains difficult due to the reliance on invasive procedures that carry risks such as arterial damage and bleeding. With advancements in big data and deep learning technologies, there is growing potential for automated and accurate diagnoses through electrocardiography (ECG). This study introduces a deep learning algorithm designed specifically for diagnosing VA using ECG data. As a non-invasive approach, it offers a cost-effective and accessible solution for patients with suspected VA. To the best of our knowledge, this is the first study to apply a deep learning model solely to ECG data for VA diagnosis.

The majority of color histogram equalization methods do not yield uniform histogram in gray scale. After converting a color histogram equalized image into gray scale, the contrast of the converted image is worse than that of an 1-D gray scale histogram equalized image. We propose a novel 3-D color histogram equalization method that produces uniform distribution in gray scale histogram by defining a new cumulative probability density function in 3-D color space. Test results with natural and synthetic images are presented to compare and analyze various color histogram equalization algorithms based upon 3-D color histograms. We also present theoretical analysis for nonideal performance of existing methods.