Ultrasound assessment of cervical lymph node metastasis in papillary thyroid carcinoma is crucial but suffers from inter-observer variability. Existing deep learning solutions typically use separate single-task pipelines, resulting in parameter redundancy. We propose MTL-LymphNet, a lightweight, backbone-agnostic multi-task learning framework that simultaneously segments lymph nodes and classifies metastasis within a single forward pass, evaluated across 3 backbones and 4 loss weighting strategies. On 338 images from 2 independent clinical centers, MTL-LymphNet achieved an internal cross-validation Dice coefficient of 0.891 ± 0.022 and AUC of 0.948 ± 0.027. Bidirectional external validation revealed substantial domain shift, with cross-center Dice of 0.865 and 0.779, and AUC of 0.733 and 0.830. The multi-task configuration offered a favorable efficiency–performance trade-off (∆Dice = Δ , − 0 . 014 , P < .0001) for 40% fewer parameters than separate models. No backbone or weighting strategy consistently dominated, and calibration degraded substantially under domain shift (ECE: 0.049 internally vs 0.178-0.304 externally), requiring center-specific recalibration before clinical use. To our knowledge, this is the first MTL framework for cervical lymph node analysis with bidirectional multi-center validation and systematic ablation. With a compact size of 22.6 to 29.1 MB and CPU inference of 81 to 131 ms, MTL-LymphNet demonstrates technical feasibility as a CAD tool warranting prospective validation.
Ovarian cyst identification is a critical task in gynecological diagnostics, aiming to distinguish between benign and malignant cysts for timely and appropriate treatment. Conventional models often use hand-engineered features, but this approach fails to capture the rich variability and detailed complexity found in ultrasound and MRI images. Additionally, classical approaches are prone to lower accuracy and poor generalizability across diverse patient populations and imaging conditions. Thus, this paper introduces a Quantum-inspired Levenberg-Marquardt Network (Quantum-LMNet) for ovarian cyst detection. Initially, the preprocessing is done using the Contra-Harmonic Mean Filter. The segmentation of cyst regions is carried out with AuxSegNet, employing multiple loss functions including multi-label soft margin loss, Dice loss, and focal loss. Features are subsequently derived from the Discrete Curvelet Transform, Local Quantized Patterns (LQP), and statistical metrics. Finally, ovarian cysts are identified using the proposed Quantum-LMNet, which combines the Levenberg-Marquardt Network (LMNet), with Quantum-Inspired Convolutional Neural Networks (QuCNet). The Quantum-LMNet also showed notable enhancement across several evaluation metrics, including Accuracy, True Positive Rate (TPR), True Negative Rate (TNR), False Omission Rate (FOR), Matthews Correlation Coefficient (MCC), and Cohen's Kappa, achieving values of 96.157%, 95.720%, 97.007%, 0.099, 0.970, and 95.359%, respectively.
Accurate standard-view classification is essential for pediatric echocardiographic image analysis and downstream automated interpretation. This task remains challenging because discriminative view information is often encoded in subtle chamber configurations, outflow-tract morphology, and weak anatomical boundaries, whereas conventional classifiers may underuse shallow and intermediate representations that preserve spatial structure. We propose PVTv2-ASEF, a structure-aware framework for 4-class pediatric echocardiographic standard-view classification. The framework introduces an Adaptive Structural Enhancement Module that performs residual input-side conditioning through channel recalibration, local convolutional mixing, multi-scale structural modeling, and input-dependent branch weighting. It further employs a Dual Auxiliary Fusion Head to transform Stage 2 and Stage 3 representations into class-level evidence and fuse them with the final-stage logits during inference. PVTv2-ASEF was evaluated on a private pediatric ventricular septal defect echocardiography dataset comprising 4 standard views under repeated patient-disjoint evaluation, with macro-F1 used as the primary class-balanced metric. Compared with PVTv2-B2, the proposed framework improved macro-F1 by 0.093 on the private dataset and by 0.023 in cross-task evaluation on FETAL_PLANES_DB. These results support the utility of coordinated input-side structural enhancement and intermediate logit fusion for ultrasound view and plane classification.
To compare pykan and Lightweight KAN with MLP and Linear classifier heads for focal liver lesion classification on B-mode ultrasound, with emphasis on diagnostic sensitivity, interpretability and computational cost. A frozen ResNet18 feature extractor was combined with 4 classifier heads and evaluated on 735 annotated liver ultrasound images (Benign, Malignant, Normal) using 5-fold stratified cross-validation. Weighted cross-entropy was applied to address class imbalance. Performance was assessed using Accuracy, F1-Macro, F1-Weighted, AUROC, class-wise recall, malignant-to-normal misclassification count, inference latency, and pykan-specific spline activeness analysis. Overall metrics were similar across models, and Wilcoxon Signed-Rank testing showed no statistically significant pairwise differences ( p > .05). However, clinically relevant class-level differences emerged. pykan achieved the highest Benign recall (65.5%) and the lowest malignant-to-normal error rate (5/435, 1.1%). Lightweight KAN achieved the highest Malignant recall (82.1%) and markedly faster inference (0.69 ms/image) than pykan (4183.67 ms/image). Aggregate metrics alone do not capture clinically meaningful trade-offs among KAN implementations. Lightweight KAN is attractive for rapid screening, whereas pykan shows a directional trend toward fewer malignant-to-normal errors and a more interpretable profile, though this trend did not reach statistical significance and requires confirmation in larger studies.
Breast ultrasound (BUS) images are commonly degraded by speckle noise, intensity inhomogeneity, acoustic shadowing, and low contrast, while breast tumors often exhibit irregular morphology, heterogeneous echogenicity, and blurred boundaries. These factors make accurate lesion segmentation difficult. We propose SFENet, an end-to-end spatial-frequency dual-branch image quality enhancement network for BUS tumor segmentation. Built on a U-shaped segmentation framework, SFENet jointly models lesion morphology, global spectral responses, and local image-quality degradation through a wavelet-frequency deformable spatial (WFDS) block. Specifically, the lesion-adaptive deformable scanning (LADS) module captures irregular lesion morphology and long-range spatial dependencies; the unified spectral filtering attention (USFA) module performs learnable Fourier-domain spectral recalibration; and the wavelet image enhancement module (WIEM) conducts local brightness correction and edge-aware contrast enhancement through wavelet sub-band decomposition. Because the enhancement pathway is optimized under segmentation supervision, it tends to strengthen lesion-relevant structural cues rather than generic visual quality alone. Experiments on the public BUSI data set show that SFENet achieves a Dice coefficient of 81.3% and a Jaccard index of 73.5%, outperforming the representative CNN-, Transformer-, and Mamba-based segmentation models included in this study. Qualitative results and Grad-CAM visualizations further suggest that SFENet produces more stable lesion contours in low-contrast, heterogeneous, and boundary-ambiguous cases. These results indicate that joint spatial-frequency modeling with task-driven wavelet-domain enhancement is an effective strategy for automatic BUS tumor segmentation.
