OBJECTIVE:To develop and validate a deep learning model integrating multi-modal ultrasound information from B-mode ultrasound (BMUS) and strain elastography (SE) for accurate preoperative prediction of central lymph node metastasis (CLNM) in patients with papillary thyroid carcinoma (PTC). METHODS:In this retrospective multi-center study, 568 patients with PTC from 4 hospitals were enrolled and divided into a training set (n = 400), an internal validation set (n = 100) and an external test set (n = 68). BMUS and SE images were collected for each patient. After image pre-processing, deep features were extracted from each modality using EfficientNet-B4 and fused for CLNM prediction. Model performance was evaluated using the area under the curve (AUC), accuracy, sensitivity and specificity. Calibration and decision curve analysis were further performed. Diagnostic performance was further compared with radiologists of different experience levels, with and without artificial intelligence assistance. RESULTS:In the internal validation set, the multi-modal fusion model achieved an AUC of 0.929, outperforming the BMUS-only (AUC = 0.843) and SE-only (AUC = 0.875) models. In the external test set, the model maintained good generalizability with an AUC of 0.843. Calibration and decision curve analysis demonstrated good agreement and clinical net benefit. Grad-CAM visualizations demonstrate a potential alignment between the regions attended to by the model and clinical signs. The proposed model outperformed all individual radiologists (AUC range, 0.585-0.649), and artificial intelligence assistance improved radiologists' diagnostic performance (AUC increased to 0.701-0.752) and confidence. CONCLUSIONS:The proposed multi-modal deep learning model enables accurate and interpretable preoperative prediction of CLNM in PTC patients, demonstrating robust performance across centers and effective clinical decision support potential.

Qin, Xiachuan; Xiao, Weihan; Zhou, Wang; Wang, Junli; Ye, Xianjun; Ren, Tiantian; Zong, Liang; Xiu, Xiaoling; Long, Qiongxian; Yuan, Hongmei; Zhao, Junjie; Wen, Yanting; Guo, Xiaoguang; He, Fanding; Zhang, Chaoxue Author Information

This study aimed to develop a multimodal radiopathomics model utilising preoperative ultrasound (US) and fine-needle aspiration cytology (FNAC) to predict large-number cervical lymph node metastasis (CLNM) in patients with clinically lymph node-negative (cN0) papillary thyroid carcinoma (PTC). This multicentre retrospective study included patients with PTC between October 2017 and June 2024 across seven institutions. Patients were categorised based on the presence or absence of large-number CLNM in training, validation, and external testing cohorts. A clinical model was developed based on the maximum diameter of thyroid nodules. Radiomics features were extracted from US images and pathomics features were extracted from FNAC images. Feature selection was performed using univariate analysis, correlation analysis, and least absolute shrinkage and selection operator regression. Six machine learning (ML) algorithms were employed to construct radiomics, pathomics, and radiopathomics models. Predictive performance was assessed using the area under the curve (AUC), and decision curve analysis (DCA). A total of 426 patients with PTC (41.65 ± 12.47 years; 124 men) were included in this study, with 213 (50

ABSTRACT Ultrasound‐derived fat fraction (UDFF) is designed to assess the hepatic fat content quantitatively. A multicenter study that verifies the diagnostic performance of UDFF for detecting hepatic steatosis has not yet been reported. This study aimed to evaluate the performance of UDFF for diagnosing and grading hepatic steatosis. Participants referred for assessment of hepatic steatosis were prospectively recruited from eight hospitals. All participants underwent UDFF and magnetic resonance imaging proton density fat fraction (MRI‐PDFF) examinations. MRI‐PDFF was used as the reference for diagnosing hepatic steatosis. From January 2023 to July 2023, a total of 300 participants were included. The median body mass index was 25.4 kg/m2 (interquartile range: 22.7–28.1). UDFF values were positively correlated with MRI‐PDFF (R = 0.80, p < 0.001). Using MRI‐PDFF ≥ 5%, ≥ 15%, and ≥ 25% as the reference standard for detecting mild, moderate, and severe hepatic steatosis, the best cutoff values of UDFF were 7.6% (area under the receiver operating characteristic curves [AUC] = 0.90), 15.9% (AUC = 0.90), and 22.3% (AUC = 0.91), respectively. Thus, UDFF has excellent diagnostic performance in detecting and grading hepatic steatosis.

With the rapid advancement of deep learning, computer-aided diagnosis and treatment have become crucial in medicine. UNet is a widely used architecture for medical image segmentation, and various methods for improving UNet have been extensively explored. One popular approach is incorporating transformers, though their quadratic computational complexity poses challenges. Recently, State-Space Models (SSMs), exemplified by Mamba, have gained significant attention as a promising alternative due to their linear computational complexity. Another approach, neural memory Ordinary Differential Equations (nmODEs), exhibits similar principles and achieves good results. In this paper, we explore the respective strengths and weaknesses of nmODEs and SSMs and propose a novel architecture, the nmSSM decoder, which combines the advantages of both approaches. This architecture possesses powerful nonlinear representation capabilities while retaining the ability to preserve input and process global information. We construct nmSSM-UNet using the nmSSM decoder and conduct comprehensive experiments on the PH2, ISIC2018, and BU-COCO datasets to validate its effectiveness in medical image segmentation. The results demonstrate the promising application value of nmSSM-UNet. Additionally, we conducted ablation experiments to verify the effectiveness of our proposed improvements on SSMs and nmODEs.