@article{zhang2026spadicer,
  title = {Toward simultaneous pseudo-space reconstruction and cell-type deconvolution of single-cell spatial transcriptome using SpaDicer},
  author = {Zhang, Junming and Yin, Shi and Xu, Lingxi and Wu, Chen and Liu, Hui},
  journal = {Cell Reports Methods},
  publisher = {Elsevier},
  year = {2026},
  month = may,
  day = {25},
  doi = {10.1016/j.crmeth.2026.101465}
}

Single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) offer complementary insights into tissue heterogeneity. Each modality is characterized by inherent trade-offs: scRNA-seq offers high resolution but lacks spatial context, while ST retains spatial information but compromises resolution. Here, we introduce SpaDicer, an end-to-end deep-learning framework designed to bridge these gaps by unifying pseudo-spatial reconstruction of single cells and cell-type deconvolution of ST spots.

@article{yin2026alzheimer,
  title = {An Effective Alzheimer Disease Diagnosis Using Resting State fMRI Images and Broad Learning System},
  author = {Issa, Sali and Wang, Qi and Qi, Ruinan and Peng, Guangxi and Yin, Shi and Peng, Qinmu},
  journal = {Psychiatry Research: Neuroimaging},
  publisher = {Elsevier},
  year = {2026},
  month = jan,
  day = {14},
  pages = {112133},
  doi = {10.1016/j.matcom.2025.05.013}
}

In this paper, a new multiclass Alzheimer diagnosis system is proposed using Broad Learning (BL) and the combination of Local Coherence (LCOR) and Intrinsic Connectivity Contrast (ICC) parameters. A public resting state fMRI database, including healthy elderly subjects (HC), Alzheimer Disease (AD) and Mild Cognitive Impairment (MCI) patients, was chosen in this study. All rs-fMRI pre-processing and analysis were performed by CONN toolbox. Three contrast cases of AD, MCI and HC were implemented within the group-level analysis, then both LCOR and ICC parameters of the affected brain clusters were combined and collected. For diagnosis system, Broad Learning (BL) classifier is trained to classify three stages of AD, MCI and HC, respectively. Experimental results demonstrate that the proposed system achieved a high average accuracy of 99.6 percent with low training cost.

@article{zhang2026mmctnet,
  title = {MMCT-Net: a Multi-Modal Hybrid CNN-Transformer Fusion Network for Preoperative Prediction of Malignant Invasion in Pulmonary Ground-Glass Nodules},
  author = {Zhang, Jie and Mei, Kun and Feng, Zikang and Wang, Bin and Yin, Shi},
  journal = {Journal of Imaging Informatics in Medicine},
  publisher = {Springer International Publishing},
  year = {2026},
  month = jan,
  day = {5},
  pages = {1--13},
  doi = {10.1007/s10278-025-01795-x}
}

The accurate preoperative prediction of invasive adenocarcinomas in pulmonary ground-glass nodules (GGNs) is critical for determining the prevention and subsequent treatment of lung cancer. Our goal is to enhance the performance of preoperative prediction through a novel artificial intelligence model to reduce the surgical mismatch rates. We propose a multi-modal hybrid CNN-Transformer fusion network (MMCT-Net) capable of extracting multi-level deep learning features that encompass both local-to-global contextual information and 2D to 3D spatial representations for precise differentiation between preinvasive and invasive lesions. The model also incorporates an adaptive feature integration mechanism to combine these deep learning features synergistically with complementary clinical parameters and radiomics signatures. In this multicenter retrospective study, we analyzed 1-mm thin-section CT images and demonstrated the effectiveness of MMCT-Net for malignant invasion prediction.

@article{mei2025ggn,
  title = {Preoperative prediction of pulmonary ground-glass nodule infiltration status by CT-based radiomics combined with neural networks},
  author = {Mei, Kun and Feng, Zikang and Liu, Hui and Wang, Min and Ce, Chao and Yin, Shi and Zhang, Xiaoying and Wang, Bin},
  journal = {BMC Cancer},
  publisher = {BioMed Central},
  volume = {25},
  number = {1},
  pages = {659},
  year = {2025},
  month = apr,
  day = {10},
  doi = {10.1186/s12885-025-14027-w}
}

The infiltration status of pulmonary ground-glass nodules (GGNs) exhibits significant variability, demanding tailored surgical strategies and individualized postoperative adjuvant therapies. This study explored the preoperative assessment of GGN infiltration status using computed tomography (CT) imaging integrated with a neural network to enhance the precision of clinical decision-making in surgical planning and therapeutic interventions. In this multicenter retrospective study, radiomics features were extracted from CT images and combined with neural network models to improve the prediction accuracy of infiltration status, providing valuable support for individualized diagnosis and treatment of pulmonary GGNs.

@article{liu2025pdl1,
  title = {Multimodal deep learning for predicting PD-L1 biomarker and clinical immunotherapy outcomes of esophageal cancer},
  author = {Liu, Hui and Bai, Yinpu and Wang, Zhidong and Yin, Shi and Gong, Cheng and Wang, Bin},
  journal = {Frontiers in Immunology},
  publisher = {Frontiers Media SA},
  volume = {16},
  pages = {1540013},
  year = {2025},
  month = mar,
  day = {11},
  doi = {10.3389/fimmu.2025.1540013}
}

Although immune checkpoint inhibitors (ICIs) have demonstrated remarkable anti-tumor efficacy in solid tumors, only a subset of esophageal squamous cell carcinoma (ESCC) patients benefit from immunotherapy. In this study, two ESCC cohorts were established from the Third Affiliated Hospital of Soochow University in China. We developed a multimodal deep learning framework integrating pathological images, longitudinal CT scans, radiomic features, and clinical information to predict PD-L1 expression levels, immunotherapy response, and overall survival. The proposed model achieved an AUC of 0.836 for PD-L1 biomarker prediction and 0.809 for immunotherapy response prediction. Furthermore, the model demonstrated strong performance in predicting overall survival, highlighting the value of multimodal integration for precision immunotherapy in esophageal cancer.

@article{guo2025cutkidney,
  title = {A novel cross-modal data augmentation method based on contrastive unpaired translation network for kidney segmentation in ultrasound imaging},
  author = {Guo, Shuaizi and Sheng, Xiangyu and Chen, Haijie and Zhang, Jie and Peng, Qinmu and Wu, Menglin and Fischer, Katherine and Tasian, Gregory E and Fan, Yong and Yin, Shi},
  journal = {Medical Physics},
  year = {2025},
  doi = {10.1002/mp.17663}
}

Kidney ultrasound (US) image segmentation is one of the key steps in computer-aided diagnosis and treatment planning of kidney diseases. However, obtaining accurate annotations for deep learning-based segmentation remains challenging because of poor image quality and weak kidney boundaries. This paper proposes a novel cross-modal data augmentation method based on a contrastive unpaired translation network (CUT) to generate simulated labeled kidney ultrasound images and improve segmentation performance on limited labeled datasets. Experimental results demonstrate that the proposed approach effectively enhances segmentation accuracy and robustness in kidney ultrasound imaging.

@article{guo2024segcyclegan,
  title = {Cross-modal transfer learning based on an improved CycleGAN model for accurate kidney segmentation in ultrasound images},
  author = {Guo, Shuaizi and Chen, Haijie and Sheng, Xiangyu and Xiong, Yinzheng and Wu, Menglin and Fischer, Katherine and Tasian, Gregory E. and Fan, Yong and Yin, Shi},
  journal = {Ultrasound in Medicine \& Biology},
  volume = {50},
  number = {11},
  pages = {1638--1645},
  publisher = {Elsevier},
  year = {2024},
  month = nov,
  doi = {10.1016/j.ultrasmedbio.2024.07.007}
}

Deep-learning algorithms have been widely applied to automatic kidney ultrasound image segmentation. However, obtaining a large number of accurately annotated kidney labels is clinically difficult and time-consuming. To address this challenge, we propose an efficient cross-modal transfer learning framework based on an improved image-to-image translation network, Seg-CycleGAN. The proposed method generates realistic kidney ultrasound images from labeled abdominal CT images and incorporates a segmentation network to preserve anatomical structures. Experimental results demonstrate that the framework significantly improves kidney segmentation performance on limited labeled ultrasound datasets and provides a practical solution for medical image analysis.