To improve the robustness of SVM models to noise and outliers, fuzzy support vector machine (FSVM) has been proposed. However, many existing FSVM models have limitations such as their dependence on assumptions, limited optimization, and unreasonable handling of noise. To address these problems, we propose a novel approach called curriculum learning-based FSVM. Our approach employs a curriculum-learning strategy, where model initially learns easy samples to avoid noise interference and obtain a good initial solution, before proceeding to learn all samples, including hard ones. To distinguish between easy and hard samples, we introduce an adaptive density-based clustering model and it is extended to kernel feature space. Moreover, we propose a slack variable-based fuzzy membership function to evaluate the importance of samples. Additionally, our model adaptively adapts the importance of samples based on feedback during the learning process. Finally, our experimental results on popular benchmarks demonstrate that our proposed model outperforms existing competitors in terms of accuracy and robustness.
Ultrasound imaging is widely used in early pregnancy to screen for fetal brain anomalies. However, the accuracy of diagnosis can be influenced by various factors, including the sonographer’s experience and environmental conditions. To address these limitations, advanced methods are needed to enhance the efficiency and reliability of fetal anomaly screening. In this study, we propose a novel approach based on a Fetal Brain Structures Detection Network (FBStrNet) for identifying key anatomical structures in fetal brain ultrasound images. Specifically, FBStrNet builds on the YOLOv5 baseline model, incorporating a lightweight backbone to reduce model parameters, replacing the loss function, and utilizing a decoupled detection header to improve accuracy. Additionally, our method integrates prior clinical knowledge to minimize false detection rates. Experimental results demonstrate that FBStrNet outperforms state-of-the-art methods, achieving real-time detection of fetal brain anatomical structures with an inference time of just 11.5 ms. This capability enables sonographers to efficiently visualize critical anatomical features, thereby improving diagnostic precision and streamlining clinical workflows.
Accurate measurement of frontomaxillary facial (FMF) angles in prenatal ultrasound (US) scans plays a pivotal role in the screening of trisomy 21. Nevertheless, this intricate procedure heavily relies on the proficiency of the ultrasonographer and tends to be a time-intensive task. Furthermore, FMF angles are subjective when measured manually. To address this challenge, we propose a deep learning-based assisted examination framework for automatically measuring FMF angles on 2D ultrasound images. Firstly, we trained a deep learning network using 1549 fetal ultrasound images to achieve automatic and accurate segmentation of critical areas. Subsequently, a key point detection network was employed to predict the coordinates of the requisite points for calculating FMF angles. Finally, FMF angles were obtained through computational means. We employed Pearson correlation coefficients and Bland–Altman plots to assess the correlation and consistency between the model’s predictions and manual measurements. Notably, our method exhibited a mean absolute error of 2.354°, outperforming the typical standards of the junior expert. This indicates the high degree of accuracy and reliability achieved by our approach.