Abstract
Optical coherence tomography angiography (OCTA), known for its high-resolution and noninvasive imaging capability, has become a key modality for visualizing retinal vasculature. Accurate and automated segmentation of capillaries, arteries, veins, and foveal avascular zone in OCTA images is essential for quantitative analysis and disease assessment. In this paper, we propose a depth enhanced cascaded framework specifically designed for multi-class OCTA segmentation. Our method investigates the spatial distribution of vasculature in retinal images and integrates a novel self-supervised depth prediction module to learn implicit depth cues from volumetric data, thereby improving the discrimination of overlapping vascular layers. In addition, we design two topology-aware loss functions that explicitly encourage structural integrity and continuity of vessel segmentation, particularly at bifurcations and endpoints. Experiments on the OCTA-6 mm and OCTA-3 mm datasets demonstrate that our method outperforms existing state-of-the-art approaches, with mIoU gains of around 2% over prior method, IPNv2, thereby highlighting enhanced segmentation accuracy and vascular topology preservation.

Abstract
Artificial intelligence (AI), complex mathematical algorithms, is currently employed across various fields to perform tasks quickly and effectively. In this study, a novel deep-learning algorithm named (CM-Net) was developed to classify biological data obtained as images from Confocal Microscopy. The images were collected for two types of bacterial species: (Escherichia coli and Staphylococcus aureus), where the number of images was 300 for each class. To enhance the dataset, we divided each image (using the augmentation method) into a small number of images with 224 & times; 224 dimensions, resulting in a total of 7066 images for both classes. These augmented images were fed to CM-Net to ensure accurate results and avoid bias in the developed algorithms. The algorithm was trained and tested 30 times with a 5-K cross-validation for each time. The algorithm's performance was evaluated using seven metrics (accuracy, sensitivity, specificity, precision, NVA, F1-score, and MCC), where the respective results were 96.08%, 95.98%, 96.19%, 96.78%, 95.26%, 96.38%, and 92.11%, indicating the model's high accuracy and reliability. CM-Net drastically reduces bacterial identification time by automating large-scale data analysis, processing results in 8.9 min. The automation provided by CM-Net simplifies workflows, enabling non-expert workers to perform microbial identification without extensive training. The significant outcomes of applying CM-Net for bacterial identification revolve around its transformative impact on data analysis's speed, efficiency, and accuracy, making advanced analysis accessible to non-experts while minimizing human error.

Abstract

Abstract

Abstract
This paper presents a smartphone application that supports Shotokan Karate training by analysing posture and providing real-time feedback. The app evaluates three fundamental stances (Zenkutsu Dachi, Kokutsu Dachi, and Kiba Dachi) using Google ML Kit Pose Detection to extract body landmarks and compute joint-angle and alignment features, including proxy indicators of weight shift. The app also includes conditioning exercises (squats and push-ups) and a reflex-oriented interaction task. Results from a single-participant pilot are reported as feasibility evidence only and should not be generalised. A larger validation study with at least 30 practitioners across three skill levels (beginner, intermediate, advanced) is required, together with power analysis and reliability assessment, before broader conclusions can be drawn.

Abstract
This article investigates the ability of large language models (LLMs) to evaluate semantic relations between word pairs by examining their alignment with human-generated semantic ratings. Semantic relations represent the degree of connection (e.g., relatedness or similarity) between linguistic elements and are traditionally validated against human-annotated datasets. Due to the challenges of building such datasets and recent progress in LLMs' capacity to model humanlike understanding, we explore whether LLMs can serve as reliable substitutes for traditional human ratings. We conducted experiments using multiple LLMs from OpenAI, Google, Mistral, and Anthropic, evaluating their performance across diverse English and Portuguese semantic relations datasets. We included in the analysis PAP900, a recently published dataset of semantic relations in Portuguese, to examine the influence of prior exposure to the dataset on LLM training. The results show that the LLM predictions correlate strongly with human ratings. The findings reveal the potential of LLMs to supplement or replace traditional semantic measure algorithms and crowd-sourced human annotations in semantic tasks.