Cephalometric analysis plays a vital role in the domains of orthodontics, where it is consistently employed. The primary procedure involved in this analysis entails identifying craniofacial landmarks on lateral cephalograms. These landmarks yield diagnostic insights into a patient’s craniofacial state and influence the decisions made regarding treatment planning. Owing to the variability in anatomical structures among individuals and the quality of X-ray imaging, achieving precise and consistent identification of landmarks within high-resolution lateral cephalograms poses a considerable challenge. Automatically and accurately locating these landmarks is a challenging issue because X-ray images possess high resolution, and the markers exhibit distinct clustering characteristics on specific organs, Based on this observation, we present a new vision transformer based approach for accurate anatomical Facial Landmark detection named FaLdViT. The proposed method comprises a target detection model designed to differentiate between the regions of the ears, eyes, and mouth in X-ray images, along with a hybrid landmarks detection model. In our network model FaLdViT, within a precision range of 2.0mm, the model achieved 77.36% average accuracy for each landmark point. which is the acceptable precision range in clinical practice, and the code is about to be open-sourced.

Open-source software (OSS) enables users to access, modify, distribute software based on open-source licenses, serving as vital digital infrastructure. Notably, GitHub stands out as a prominent OSS community, with 94 million developers engaged in projects by 2022. However, accurately assessing women’s contributions in OSS encounters challenges due to limited gender data. To address this, we propose an innovative method that employs the Large-Language-Model (LLM), ChatLM2. This LLM-based approach allows cross-lingual analysis of women’s involvement and quantitatively assesses their impact on OSS projects. The study aims to uncover gender disparities and encourage greater participation of female developers in the open-source realm. The article is structured with sections on research methods, design, LLM-based gender detection, women’s participation, impact assessment, implications, and future research.

Parkinsons disease (PD) is a brain disorder that causes unintended or uncontrollable movements. The diagnosis of the disease mainly relies on clinical test rather than a definite medical test, and the diagnostic accuracy is only about 80%. Thus, an efficient and explainable automatic PD diagnosis system is valuable for supporting clinicians with more robust diagnostic decision-making. We present a novel way to classify Parkinsons disease from RGB videos, which is contactless and accurate. Our new algorithm is SimpleHandFormer, which only requires non-intrusive RGB-video recordings of the candidates as input. Our algorithm is motivated by the observation that tremors (shaking movement) will cause the hand keypoint’s location changes, and focusing on the change of the keypoints would improve the performance. For the first time, we propose to use a novel frame–keypoint different attention to effectively detect tremors in the human hands. This design aids in improving both binary classification performance and algorithm explanation. Experimental results show that our system outperforms state-of-the-arts by achieving a balanced accuracy of 93.9% and an F1-score of 92.6% in classifying Parkinson’s Disease.

Existing studies in both psychology and software engineering have shown the importance of emotions in complex learning and programming tasks. For students who are learning to program, rich emotions are experienced which can provide valuable feedback to their teachers. To accurately model stu-dents' emotions, this paper adopts the well-recognized model of emotions during complex learning that involves four states: engaged, confused, frustrated, and bored. To perform continuous estimation of students' emotions in a non-intrusive manner, this paper proposes to track students' programming behavior and estimate their corresponding emotional states. Compare to the existing approaches on acquiring the students' emotional states with self-reports or bio-sensors, the proposed approach is more feasible in conducting real-world, and large-scale studies for not requiring extensive human interventions or additional devices. Evaluated using data collected from a real-world course project, the proposed approach is showed to be promising for achieving an estimation accuracy of 72.06 % for the above four emotional states. As an enabling technology, the proposed is potentially useful in supporting many applications and improve the quality of programming education in computer science.

Flow is a holistic description of people's optimal experiences during creative activities that can be characterized as being totally concentrated on, and actively involved in the task, enjoying the process of creation, and achieving a balance between one's skill and the task's challenge. Understanding software developers' flow states has attracted an increasing attention in both research and practice because of the strong link between being in flow and achieving good performance. In this paper, we study the problem of tracking and recognizing developers' flow states by tracing their computer interactions including activities of using the keyboard, mouse, IDE functions, and switching application windows. Compared to the traditional approaches that rely on self-reports or wearable sensors, a major advantage of the proposed approach is being non-invasive for not requiring any additional efforts from the developers after the training phase is completed, which is important because the developers' flow states can easily be interrupted by external interferences. Based on the captured interaction traces, we represent the developers' activities with extensive features, and propose to address the flow state recognition problem using machine learning technologies. And a hierarchical recognition model is built following the multi-dimensional construct of the flow concept, which is interpretable and effective. We develop a prototype system and conduct a 17-day field study in a medium-sized IT company in China to collect real-world data. The results show that our approach is effective by achieving the highest recognition accuracy of 92.6%, and efficient for performing real-time recognition.