Human posture recognition is one of the most challenging tasks due to the variation in human appearance, changes in the background and illumination, additional noise in the frame, and diverse characteristics and amount of data generated. Aside from these, generating a high configuration for recognition of human body parts, occlusion, nearly identical parts of the body, variations of colors due to clothing, and other various factors make this task one of the hardest in computer vision. Therefore, these studies require high-computing devices and machines that could handle the computational load of this task. This study used a small-scale convolutional neural network and a smartphone built-in camera to recognize proper and improper sitting posture in a work-from-home setup. Aside from the recognition of body points, this study also utilized points’ distances and angles to help in recognition. Overall, the study was able to develop two objective datasets capturing the left and right side of the participants with the supervision and guidance of licensed physical therapists. The study shows accuracies of 85.18% and 92.07%, and kappas of 0.691 and 0.838, respectively. The system was developed, implemented, and tested in a work-from-home environment.

Uttar Pradesh (UP), the most populous state in India, has about 36 million people aged 50 years or older, spread across more than 100,000 villages. Among them, an estimated 3.5 million suffer from visual impairments, including blindness due to untreated cataracts. To achieve cataract backlog-free status, UP is required to screen this population at the community level and provide treatment to those suffering from cataracts. We envisioned an AI-powered primary screening app utilizing eye images, deployable to frontline health workers for community-level screening. This paper outlines insights gained from developing the AI mobile app "Roshni" for cataract screening. The AI-based cataract classification model was developed using 13,633 eye images and finalized after three stages of experiments, detecting cataracts in images focused on the eye, iris, and pupil. Overall, 155 experiments were conducted using multiple deep learning algorithms, including ResNet50, ResNet101, YOLOv5, EfficientNetV2, and InceptionV3. We established a minimum threshold of 90% specificity and sensitivity to ensure the algorithm's suitability for field use. The cataract detection model for eye-focused images achieved 51.9% sensitivity and 87.6% specificity, while the model for iris-focused images, using a good/bad iris filter, achieved 52.4% sensitivity and 93.3% specificity. The classification model for segmented-pupil images, employing a good/bad pupil filter with UNet-based semantic segmentation model and EfficientNetV2, yielded 96% sensitivity and 97% specificity. Field testing with 302 beneficiaries (604 images) showed an overall sensitivity of 86.6%, specificity of 93.3%, positive predictive value of 58.4%, and negative predictive value of 98.5%. This paper details the development of an AI mobile app designed to facilitate community screening for cataracts by frontline health workers.

The development of science and technology has made significant breakthroughs in the field of image recognition, and has applied to fields such as medicine, education, and transportation. Smart education will apply deep learning based recognition methods to student classroom behavior detection, improving teaching quality and classroom management efficiency. This article first introduces the current research status of machine learning in the field of behavior detection; Then, the widely used deep learning based detection algorithms were introduced in detail; Finally, a comparison and summary of classroom behavior detection algorithms were conducted.

Retinopathy of prematurity (ROP) is one of the main causes of blindness in children. Timely prevention and treatment can effectively control its progression. Although convolutional neural networks (CNNs) have been widely studied in the field of ROP detection, limitations such as parameter redundancy, spatial information loss, and limited interpretability persist. To address these issues, we proposed LDA-UNet, a lightweight CNN model that integrates depth separable convolution, multi-scale dilated convolution, and residual channel-spatial attention mechanisms. Our model significantly reduces parameters (by 85.4% compared to standard UNet) while alleviating the lack of spatial information loss during the encoding stage. The Dice coefficient is 0.927, which exceeds both the current typical lightweight UNet models and the Transformer-based hybrid structure TransUNet. By combining LDA-UNet with a contour detection algorithm, we introduce an object visualization model that precisely locates lesion areas and delineates contours, enhancing interpretability. Our approach outperforms YOLOv5s in mean Intersection over Union (mIoU) by 17.5% and object detection accuracy by 12.5%, underscoring its efficacy in ROP visualization. This work presents a robust, lightweight solution for ROP diagnosis, facilitating clinical decision-making and improving patient outcomes. Codes are available at https://github.com/honghaolu/ROP_Visualizaiton or https://doi.org/10.5281/zenodo.13922434.

Rice false smut is an important fungal infection in the rice panicle stage, which occurs only in the panicle. Rice yield and quality will be seriously threatened after the occurrence of panicle disease. Early identification of disease is very important for precise prevention and control. However, in the actual field environments, complex light changes, the dense distribution of small disease spots, panicle overlapping shading, and other factors often result in the semantic attenuation of key discriminant information in the stage of visual feature extraction, which has brought great challenges to the early detection and prevention of the disease. To resolve the above problems, this study introduces a rice false smut detection model derived from an improved YOLOv11 framework, named Rice-Smut, to bolster the resilience and stability of the network regarding the identification of rice false smut disease under complex field backgrounds. Firstly, in order to enhance the feature capture capabilities for multi-scale and densely distributed lesions, the C3SC backbone feature extraction network combining the SCConv block is integrated. This architecture can significantly suppress the spatial and channel redundancy and augment the precise characterization of the texture features of the lesion. Then, the C2PSA-SE attention module is introduced to effectively filter the background interference and improve the precise positioning of dense small targets. Finally, to address the irregular structure of rice false smut lesions, the GIoU loss function serves as a substitute for the conventional CIoU, which enhances the network’s proficiency in locating the irregular shape lesions. Experimental outcomes revealed that the Rice-Smut model yielded a precision of 79.3% and mAP@50 of 75.3%, which represented a 7.6 and 4.5 percentage point improvement over the baseline model YOLOv11. The model requires 2.41M parameters, with a model size of 4.9MB, which results in low computational complexity. The preliminary validation on mobile platforms shows that the method is viable for the potential to be applied to the real-time field detection and disease monitoring of rice false smut, and can provide support for disease control decision-making and field management.

[Objective] This review summarizes the current state of automated dental implant planning based on deep learning (DL) and Cone-Beam Computed Tomography (CBCT) imaging, proposing a novel closed-loop planning paradigm driven by clinical outcome feedback. [Methods] The necessity of digital transformation for enhancing implant precision and biomechanical stability is first analyzed. Technical evolution is then reviewed: anatomical segmentation has transitioned from early level-set methods to instance segmentation and attention-based U-Net for nerve canal tracking. Applications include implant system identification and metal artifact reduction (MAR) using Generative Adversarial Networks (GANs). Regarding core algorithms, end-to-end regression models like ImplantFormer are analyzed, highlighting the advantages of Transformer architectures in global spatial modeling. [Results] Addressing the limitation of "open-loop" models that rely heavily on subjective expert annotations, this project proposes a "closed-loop feedback" model. By registering pre- and postoperative images, successful clinical outcomes are mapped back to the planning stage. This allows the AI to learn and pre-compensate for systemic deviations, such as drill deflection during execution, while overcoming the challenge of inter-observer variability. [Conclusion] AI is driving the shift in dental implantology from "assistive tools" to "autonomous decision systems." Future integration of "planning-execution-feedback" cycles with surgical robotics will facilitate a paradigm shift from static digital design to outcome-driven, zero-deviation automated planning.