Introduction: Rheumatoid arthritis (RA) is a systemic autoimmune disease; early diagnosis and treatment are crucial for its management. Currently, the modified total Sharp score (mTSS) is widely used as a scoring system for RA. The standard screening process for assessing mTSS is tedious and time-consuming. Therefore, developing an efficient mTSS automatic localization and classification system is of urgent need for RA diagnosis. Current research mostly focuses on the classification of finger joints. Due to the insufficient detection ability of the carpal part, these methods cannot cover all the diagnostic needs of mTSS. Method: We propose not only an automatic label system leveraging the You Only Look Once (YOLO) model to detect the regions of joints of the two hands in hand X-ray images for preprocessing of joint space narrowing in mTSS, but also a joint classification model depending on the severity of the mTSS-based disease. In the image processing of the data, the window level is used to simulate the processing method of the clinician, the training data of the different carpal and finger bones of human vision are separated and integrated, and the resolution is increased or decreased to observe the changes in the accuracy of the model. Results: Integrated data proved to be beneficial. The mean average precision of the proposed model in joint detection of joint space narrowing reached 0.92, and the precision, recall, and F1 score all reached 0.94 to 0.95. For the joint classification, the average accuracy was 0.88, and the accuracy of severe, mild, and healthy reached 0.91, 0.79, and 0.9, respectively. Conclusions: The proposed model is feasible and efficient. It could be helpful for subsequent research on computer-aided diagnosis in RA. We suggest that applying the one-hand X-ray imaging protocol can improve the accuracy of mTSS classification model in determining mild disease if it is used in clinical practice.

Induced pluripotent stem cells (iPSCs) can be differentiated into mesenchymal stem cells (iPSC-MSCs), retinal ganglion cells (iPSC-RGCs), and retinal pigmental epithelium cells (iPSC-RPEs) to meet the demand of regeneration medicine. Since the production of iPSCs and iPSC-derived cell lineages generally requires massive and time-consuming laboratory work, artificial intelligence (AI)-assisted approach that can facilitate the cell classification and recognize the cell differentiation degree is of critical demand. In this study, we propose the multi-slice tensor model, a modified convolutional neural network (CNN) designed to classify iPSC-derived cells and evaluate the differentiation efficiency of iPSC-RPEs. We removed the fully connected layers and projected the features using principle component analysis (PCA), and subsequently classified iPSC-RPEs according to various differentiation degree. With the assistance of the support vector machine (SVM), this model further showed capabilities to classify iPSCs, iPSC-MSCs, iPSC-RPEs, and iPSC-RGCs with an accuracy of 97.8%. In addition, the proposed model accurately recognized the differentiation of iPSC-RPEs and showed the potential to identify the candidate cells with ideal features and simultaneously exclude cells with immature/abnormal phenotypes. This rapid screening/classification system may facilitate the translation of iPSC-based technologies into clinical uses, such as cell transplantation therapy.

Purpose: Rheumatoid arthritis (RA) is a chronic autoimmune disease whose accurate diagnosis relies on clinical symptoms, serological markers, and imaging studies, particularly X-ray evaluations of joint damage. A widely adopted metric for evaluating RA severity in hand and foot radiographs, the modified Total Sharp Score, quantifies joint erosion and joint space narrowing. However, manual scoring is labor-intensive, time-consuming, and prone to variability, limiting its reliability and consistency in clinical practice. Methods: This study introduces a deep learning–based approach augmented with an attention mechanism to automate joint localization and damage classification in hand X-rays of patients with RA. The You Only Look Once v7 model was used for joint localization, incorporating performance enhancements via window-level transformation and the Distance Intersection over Union algorithm. Results: The model achieved a localization accuracy of 99.66% mAP@0.50, significantly enhancing the detection precision. This model outperformed the vision transformer in scenarios involving limited data and small feature regions, achieving an overall classification accuracy of 88%. Conclusion: The proposed method improved feature learning and classification, especially with limited data, and the attention mechanisms significantly enhanced the performance of the RA automatic diagnosis system. Data imbalance was effectively addressed via a modified EfficientNetV2 model with an integrated attention mechanism for joint damage classification.

Floating debris in aquatic environments poses ecological risks, necessitating prompt detection to prevent its spread into oceans, where recovery is challenging. This letter introduces an optimized YOLOv4-based detection framework tailored for small-scale floating debris on embedded systems, specifically for the Raspberry Pi 4 with integrated camera modules. Key innovations include a modified REGP pooling layer, enhanced Spatial Pyramid Pooling (SPP), and reduced detection heads, boosting mean Average Precision (mAP) by 7.91% on the FloW dataset while reducing parameters by 26.35% compared to the baseline model. These improvements enhance computational efficiency, achieving real-time performance at 15 fps with 2.8W power consumption, making it ideal for large-scale environmental monitoring.

Purpose: In this paper, we propose an open-source deep learning-based computer-aided diagnosis system for breast ultrasound images based on the Breast Imaging Reporting and Data System (BI-RADS). Methods: Our dataset with 8,026 region-of-interest images preprocessed with ten times data augmentation. We compared the classification performance of VGG-16, ResNet-50, and DenseNet-121 and two ensemble methods integrated the single models. Results: The ensemble model achieved the best performance, with 81.8% accuracy. Our results show that our model is performant enough to classify Category 2 and Category 4/5 lesions, and data augmentation can improve the classification performance of Category 3. Conclusion: Our main contribution is to classify breast ultrasound lesions into BI-RADS assessment classes that place more emphasis on adhering to the BI-RADS medical suggestions including recommending routine follow-up tracing (Category 2), short-term follow-up tracing (Category 3) and biopsies (Category 4/5).