Multi-Task Swin Transformer Framework for Gastrointestinal Cancer Pathology

Qing-Chun Feng's research dataset, last updated April 2026, contains approximately 99,000 H&E-stained histopathological image patches from the GasHisSDB and GCHTID datasets. The data underpins a study developing a multi-task deep learning framework for classification and segmentation of tumor and microenvironment structures in gastrointestinal cancer. The framework achieved a classification F1-score of 0.938 and a segmentation Dice coefficient of 0.839 on the test set.

Use Cases

• Training multi-task deep learning models for joint classification and segmentation based on histopathological image patches.
• Benchmarking model robustness and cross-dataset generalization based on the described transfer from GCHTID to GasHisSDB.
• Evaluating model explainability methods like Grad-CAM and LRP based on the described expert-model agreement rate (EMAR) of 0.864.
• Testing model stability under staining perturbations based on the reported Dice coefficient decrease of only 2.4%.

Strengths

• Approximately 99,000 H&E-stained histopathology image patches from two established datasets (GasHisSDB and GCHTID).
• Model performance is quantitatively validated with a classification F1-score of 0.938 ± 0.007 and segmentation Dice of 0.839 ± 0.009.
• Explainability is assessed with multiple methods (Grad-CAM, Score-CAM, LRP) and validated by pathology experts, achieving an expert-model agreement rate (EMAR) of 0.864.

Limitations

• The dataset is very small at 13.2 KB, suggesting it contains documentation or summary results rather than the primary image data.
• Row count and column-level documentation are unknown, limiting suitability assessment for direct data reuse.
• The validation was conducted under controlled benchmark conditions using patch-level data; clinical feasibility requires further verification on whole-slide images.

Provenance

Source

Qing-Chun Feng via figshare.

Collection Method

Images were collected from the GasHisSDB and GCHTID datasets, with preprocessing including color normalization and quality control.

Freshness

Last updated 2026-04-21 04:11:11.