Accurate segmentation of small anatomical structures like the pubic symphysis in intrapartum ultrasound images is critical for clinical diagnosis, yet remains challenging due to inherent noise, speckle artifacts, low target-background contrast, and blurred boundaries. Established methods often fail to capture long-range dependencies while preserving the spatial continuity of small targets, leading to inaccurate boundary localization. To address these issues, we propose an Edge-Aware Network with Large Kernel Attention (EAN_LKA). Our approach integrates three novel components: a Spatial-Continuous Encoder which capture long-range dependencies via shift operations in order to avoid structural fragmentation caused by traditional patch-based mechanisms and to maintain pixel-level boundary coherence; a Cross-Scale Fusion Attention module which enhances semantic discriminability between target and background by refining multi-scale features between the encoder and decoder in order to effectively mitigate low-contrast boundary ambiguity; and a Large Kernel Attention Decoder which leverages extensive contextual perception to suppress noise interference while guiding detail-preserving boundary reconstruction. Experiments on two public MICCAI challenge datasets demonstrate that the EAN_LKA model significantly outperforms existing methods in small target segmentation tasks, with an average Dice score improvement of 1.18 % and a relative reduction in average surface distance of$\mathbf{3. 8 7 \%}$, fully validating its overall superiority.

Cardiotocography (CTG) is essential for monitoring fetal health. Current deep learning applications in this field face two challenges: 1) the scarcity of annotated CTG data; 2) the difficulty in simultaneously capturing the coupled dependencies within multivariate time-series CTG. To bridge this gap, we propose MCTG: a multimodal self-supervised contrastive learning framework. For positive samples, we integrate an augmentation method based on frequency shapelets. These short frequency-domain subsequences capture class-specific information, preserving the most discriminative components of CTG signals. Additionally, we construct a multimodal learning framework that integrates series-image modalities to obtain complementary information between different modalities. This approach compensates for the inherent limitations of single time series modality networks in terms of complex cross-variable dependencies. Specifically, we encode series into images through three-channel encoding (frequency, wavelet, and time) to capture global long-term features, cross-scale features, and temporal texture details. Concurrently, multi-scale convolution integrates each variable dimension of the time series into a single image, enhancing the model’s ability to capture dependency couplings between variables. Experimental results demonstrate that MCTG effectively learns crucial feature representations from CTG data and achieves state-of-the-art performance in fetal distress prediction. The codes are available at https://github.com/Ladyfish030/MCTG.