Delayed treatment of infantile retinal disease can reduce its effectiveness and may cause severe and irreversible damage. Automated diagnosis of infant retinal diseases faces challenges including subtle early lesions, diverse clinical phenotypes, imaging variations, and imbalanced data. To address these, which cannot be well addressed by existing general foundation models, we propose structure-aware bidirectional proxy interaction network (SABPI-Net) in a universal learning framework. SABPI-Net incorporates a high-frequency mapping branch, and employs a proposed proxy interaction attention module to enable effective interaction between its trunk feature encoding branch and the high-frequency mapping branch, thereby facilitating enhanced perception of retinal detail structures. Domain-agnostic embedding space self-matching, guided by a memory-bank low-frequency component replacement strategy, promotes domain-invariant learning and consistent model performance under diverse image styles. Finally, the tail-aware feature fusion strategy for fine-tuning further enhances the model's diagnostic sensitivity to tailed diseases. In this study, three classification tasks related to infant retinal diseases are implemented on the largest clinical infant retina dataset to date, covering 19 infant retinal diseases or normal conditions. SABPI-Net achieves superior performance compared to 13 SOTA methods, with 95.32% accuracy on mainstream clinical tasks, 73.58% on ROP five-stage classification, and 84.25% on multi-disease classification, representing improvements of 1.57%, 1.88%, and 4.71% respectively over the best competing methods. Extensive experiments demonstrate the effectiveness and superiority of SABPI-Net in diagnosing infant retinal diseases.

Accurate segmentation of breast lesions is vital for diagnosing breast diseases. Masked image modeling (MIM) with random masking performs well in self-supervised learning but struggles in breast ultrasound segmentation due to (1) ambiguous representations from similar intensities near lesion boundaries and (2) a bias toward irrelevant regions. We propose UA-MAE, an uncertainty-aware masked autoencoder that uses pixel-wise uncertainty maps to dynamically select masking patches, prioritizing boundaries and morphologically relevant lesion areas. Experiments on two public datasets for pre-training and three for fine-tuning show UA-MAE outperforming four state-of-theart SSL methods and two supervised approaches in segmentation accuracy across diverse breast ultrasound images. The code is available at https://github.com/yXiangXiong/UA-MAE.

Conventional one-stop authentication of a computer terminal takes place at a user's initial sign-on. In contrast, continuous authentication protects against the case where an intruder takes over an authenticated terminal or simply has access to sign-on credentials. Behavioral biometrics has had some success in providing continuous authentication without requiring additional hardware. However, further advancement requires benchmarking existing algorithms against large, shared datasets. To this end, we provide a novel large dataset that captures not only keystrokes, but also mouse events and active programs. Our dataset is collected using passive logging software to monitor user interactions with the mouse, keyboard, and software programs. Data was collected from 103 users in a completely uncontrolled, natural setting, over a time span of 2.5 years. We apply Gunetti & Picardi's algorithm, a state-of-the-art algorithm in free text keystroke dynamics, as an initial benchmarkfor the new dataset.

Keystroke authentication can help significantly improve computer security by hardening passwords or offering active, continuous authentication. Over the years, many keystroke authentication algorithms have been reported to produce promising results. However, these results are tested on proprietary datasets with varying numbers of subjects and amounts of text, making it difficult to compare and improve the state of art. We describe a new dataset that we have developed with the goal to serve as a shared common testbed to enable future improvements. The new dataset includes keystroke data for short pass-phrases, fixed text (transcription of long proses), and free text. It also includes video of a subject's facial expression and hand movement during the data collection sessions, allowing for a deeper understanding of why an algorithm works the way it does, for example, by finding out whether a subject is a touchtypist or not. As a baseline for benchmarking, we also include the results of replicating two existing algorithms using the new dataset.