Back to all papers

SiCLIP: An explainable multimodal framework for silicosis diagnosis.

May 23, 2026pubmed logopapers

Authors

Le D,Nguyen T,Nguyen H,Nguyen A,Tran C,Pham C

Affiliations (6)

  • Posts and Telecommunications Institute of Technology, Hanoi, Viet Nam. Electronic address: [email protected].
  • Thai Nguyen University of Medicine and Pharmacy, Thai Nguyen, Viet Nam. Electronic address: [email protected].
  • Posts and Telecommunications Institute of Technology, Hanoi, Viet Nam. Electronic address: [email protected].
  • Hanoi Medical University, Hanoi, Viet Nam. Electronic address: [email protected].
  • Posts and Telecommunications Institute of Technology, Hanoi, Viet Nam. Electronic address: [email protected].
  • Posts and Telecommunications Institute of Technology, Hanoi, Viet Nam. Electronic address: [email protected].

Abstract

Silicosis is a serious occupational lung disease caused by exposure to crystalline silica dust and remains difficult to detect early in at-risk worker populations. In this paper, we introduce the Silicosis Diagnosis Dataset (SDD), which comprises chest X-ray images and structured patient-profile information, including harmful habits and clinical symptoms. To exploit this multimodal dataset, we propose SiCLIP, a multimodal retrieval framework based on CLIP-ViT for silicosis screening and binary classification on SDD. SiCLIP learns a shared embedding space for chest X-ray images and patient profiles and performs retrieval-based aggregation for prediction. On the internally evaluated SDD benchmark, SiCLIP achieves higher accuracy and F1-score than several strong image-only deep learning baselines and the compared multimodal VLM baseline. In addition, SiCLIP provides case-based interpretability by grounding predictions in retrieved similar cases, complemented by supportive saliency visualizations. These results suggest that multimodal retrieval is a promising approach for silicosis screening support in occupationally exposed populations, while external validation remains necessary before broader clinical deployment.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.