❌

Normal view

Received β€” 27 January 2026 ⏭ Favourites

From Specialist to Generalist: Unlocking SAM's Learning Potential on Unlabeled Medical Images

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Vi Vu Β· Thanh-Huy Nguyen Β· Tien-Thinh Nguyen Β· Ba-Thinh Lam Β· Hoang-Thien Nguyen Β· Tianyang Wang Β· Xingjian Li Β· Min Xu
27 January 2026 at 13:00
arXiv:2601.17934v1 Announce Type: cross Abstract: Foundation models like the Segment Anything Model (SAM) show strong generalization, yet adapting them to medical images remains difficult due to domain shift, scarce labels, and the inability of Parameter-Efficient Fine-Tuning (PEFT) to exploit unlabeled data. While conventional models like U-Net excel in semi-supervised medical learning, their potential to assist a PEFT SAM has been largely overlooked. We introduce SC-SAM, a specialist-generalist framework where U-Net provides point-based prompts and pseudo-labels to guide SAM's adaptation, while SAM serves as a powerful generalist supervisor to regularize U-Net. This reciprocal guidance forms a bidirectional co-training loop that allows both models to effectively exploit the unlabeled data. Across prostate MRI and polyp segmentation benchmarks, our method achieves state-of-the-art results, outperforming other existing semi-supervised SAM variants and even medical foundation models like MedSAM, highlighting the value of specialist-generalist cooperation for label-efficient medical image segmentation. Our code is available at https://github.com/vnlvi2k3/SC-SAM.
Received β€” 14 January 2026 ⏭ Favourites

Imaging-anchored Multiomics in Cardiovascular Disease: Integrating Cardiac Imaging, Bulk, Single-cell, and Spatial Transcriptomics

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Minh H. N. Le Β· Tuan Vinh Β· Thanh-Huy Nguyen Β· Tao Li Β· Bao Quang Gia Le Β· Han H. Huynh Β· Monika Raj Β· Carl Yang Β· Min Xu Β· Nguyen Quoc Khanh Le
14 January 2026 at 13:00
arXiv:2601.07871v1 Announce Type: cross Abstract: Cardiovascular disease arises from interactions between inherited risk, molecular programmes, and tissue-scale remodelling that are observed clinically through imaging. Health systems now routinely generate large volumes of cardiac MRI, CT and echocardiography together with bulk, single-cell and spatial transcriptomics, yet these data are still analysed in separate pipelines. This review examines joint representations that link cardiac imaging phenotypes to transcriptomic and spatially resolved molecular states. An imaging-anchored perspective is adopted in which echocardiography, cardiac MRI and CT define a spatial phenotype of the heart, and bulk, single-cell and spatial transcriptomics provide cell-type- and location-specific molecular context. The biological and technical characteristics of these modalities are first summarised, and representation-learning strategies for each are outlined. Multimodal fusion approaches are reviewed, with emphasis on handling missing data, limited sample size, and batch effects. Finally, integrative pipelines for radiogenomics, spatial molecular alignment, and image-based prediction of gene expression are discussed, together with common failure modes, practical considerations, and open challenges. Spatial multiomics of human myocardium and atherosclerotic plaque, single-cell and spatial foundation models, and multimodal medical foundation models are collectively bringing imaging-anchored multiomics closer to large-scale cardiovascular translation.
Received β€” 13 April 2025 ⏭ Favourites

MLG2Net: Molecular Global Graph Network for Drug Response Prediction in Lung Cancer Cell Lines

βœ‡Omics In Lung
By: Thi-Oanh Tran Β· Thanh-Huy Nguyen Β· Tuan Tung Nguyen Β· Nguyen Quoc Khanh Le
10 April 2025 at 18:00

J Med Syst. 2025 Apr 10;49(1):47. doi: 10.1007/s10916-025-02182-3.

ABSTRACT

Drug response prediction (DRP) is a central task in the era of precision medicine. Over the past decade, the emergence of deep learning (DL) has greatly contributed to addressing DRP challenges. Notably, the prediction of DRP for cancer cell lines benefits significantly from data availability for model development. However, an effective predictive model is still challenging due to issues with data quality, high-dimensional data, and multi-omics data integration. In this study, we introduce MLG2Net, a deep-learning model inspired by graph neural networks designed to predict DRP in lung cancer cell lines based on pharmacogenomics data. Our model comprises two key components: drug SMILES described by local and global graph networks and cell line genomics are illustrated as a map. Our results show that MLG2Net outperforms three reference graph networks. MLG2Net performance reached a Pearson coefficient correlation ( C C p ) of 0.8616 and a root mean square error (RMSE) of 2.94e-6 in predicting drug responses for Lung Adenocarcinoma (LUAD) cell lines. Subsequent testing on the Lung Squamous Cell Carcinoma (LUSC) dataset reveals lower performance ( C C p : 0.7999, RMSE: 4.08e-6), attributed to the dataset's smaller size influencing model capacity. Moreover, we assessed the model's architecture by isolating its components, with results indicating that the global network is particularly effective in this task. In conclusion, MLG2Net exhibited promising applications in DRP for cancer cell lines, with potential advancements by incorporating larger datasets.

PMID:40208442 | DOI:10.1007/s10916-025-02182-3

MLG2Net: Molecular Global Graph Network for Drug Response Prediction in Lung Cancer Cell Lines

βœ‡(Multiomics OR Omics) AND (Lung OR gastric OR Hepatocellular)
By: Thi-Oanh Tran Β· Thanh-Huy Nguyen Β· Tuan Tung Nguyen Β· Nguyen Quoc Khanh Le
10 April 2025 at 18:00

J Med Syst. 2025 Apr 10;49(1):47. doi: 10.1007/s10916-025-02182-3.

ABSTRACT

Drug response prediction (DRP) is a central task in the era of precision medicine. Over the past decade, the emergence of deep learning (DL) has greatly contributed to addressing DRP challenges. Notably, the prediction of DRP for cancer cell lines benefits significantly from data availability for model development. However, an effective predictive model is still challenging due to issues with data quality, high-dimensional data, and multi-omics data integration. In this study, we introduce MLG2Net, a deep-learning model inspired by graph neural networks designed to predict DRP in lung cancer cell lines based on pharmacogenomics data. Our model comprises two key components: drug SMILES described by local and global graph networks and cell line genomics are illustrated as a map. Our results show that MLG2Net outperforms three reference graph networks. MLG2Net performance reached a Pearson coefficient correlation ( C C p ) of 0.8616 and a root mean square error (RMSE) of 2.94e-6 in predicting drug responses for Lung Adenocarcinoma (LUAD) cell lines. Subsequent testing on the Lung Squamous Cell Carcinoma (LUSC) dataset reveals lower performance ( C C p : 0.7999, RMSE: 4.08e-6), attributed to the dataset's smaller size influencing model capacity. Moreover, we assessed the model's architecture by isolating its components, with results indicating that the global network is particularly effective in this task. In conclusion, MLG2Net exhibited promising applications in DRP for cancer cell lines, with potential advancements by incorporating larger datasets.

PMID:40208442 | DOI:10.1007/s10916-025-02182-3

There are no more articles
❌