McGregorW/SYSMH

In-House Multimodal Medical Segmentation Dataset

Used in “Synergistic Vision-Language Reinforcement Unlocks On-Demand Analysis across Diverse Clinical Tasks”

---

📘 Overview

This dataset provides a comprehensive multimodal collection of 3D medical scans and segmentation masks, curated for research on vision–language synergy, multimodal reinforcement learning, and cross-task generalization.
It was constructed as part of the Synergistic Vision-Language Reinforcement project to enable unified analysis across diverse clinical tasks.

The dataset contains CT and MR volumes spanning over 20 disease types, including cancers, tumors, infarctions, and embolisms.
Each case is anonymized and stored in NIfTI (.nii.gz) format, with corresponding binary segmentation masks (0–1) indicating pathological regions.

---

🏥 Data Structure

InhouseDataset/
├── images/
│   ├── Acoustic_Neuroma_MR/
│   ├── Bladder_Cancer_CT/
│   ├── Brain_Tumor_MR/
│   ├── Cerebral_Hemorrhage_CT/
│   ├── Cerebral_Infarction_MR/
│   ├── Cervical_Cancer_MR/
│   └── ...
│
└── masks/
    ├── Acoustic_Neuroma/
    ├── Bladder_Cancer_CT/
    └── ...