Ejecuta cualquier Skill en Manus
con un clic
con un clic
Ejecuta cualquier Skill en Manus con un clic
Comenzar$pwd:
$ git log --oneline --stat
stars:1
forks:0
updated:30 de marzo de 2026, 14:46
SKILL.md
| name | dgcl-brain-network-construction |
| description | DGCL Brain Network Construction |
Source: arXiv:2407.18329v1 (July 2024) Utility: 0.90 Authors: Shuqiang Wang
This skill implements DGCL (Diffusion-based Graph Contrastive Learning) - an end-to-end brain network construction pipeline that addresses limitations of traditional methods: empirical dependency, weak consistency, and time-consuming processes.
Core Method:
fmri_data - Brain imaging data (ADNI, ABIDE)diffusion_model - Brain region localizationgraph_contrastive_learning - Connection optimizationtorch_geometric - Graph neural network frameworkclassification_metrics - Disease stage predictionUser: 如何用 DGCL 构建阿尔茨海默病脑网络?
Agent: DGCL 流程:
优势:
User: DGCL 能用于自闭症脑网络分析吗?
Agent: 可以!ABIDE 数据集验证:
| Dataset | Disease | DGCL Performance |
|---|---|---|
| ADNI | Alzheimer's | ✅ Superior |
| ABIDE | Autism | ✅ Superior |
步骤:
Purpose: Precisely determine spatial locations of brain regions
Method: Diffusion process avoiding subjective parameter selection
Advantage: No empirical user dependency
Purpose: Optimize brain connections by eliminating individual differences
Method:
Total Loss = Node-Graph Contrastive Loss + Classification Loss
- Node-graph contrastive: Learn discriminative node/graph features
- Classification: Disease stage prediction accuracy
fMRI Data → BRAM (Diffusion) → Initial Brain Network
↓
Graph Contrastive Learning → Optimized Connections
↓
Joint Loss Optimization → Reconstructed Brain Network
↓
Disease Classification + Important Connection Analysis
| Metric | ADNI | ABIDE |
|---|---|---|
| Disease stage prediction | Superior | Superior |
| Brain network consistency | High | High |
| Construction efficiency | End-to-end | End-to-end |
| Generalization | Strong | Strong |
Comparison vs Traditional Methods:
| Traditional | DGCL |
|---|---|
| Empirical parameter selection | ✅ Automatic (diffusion) |
| Weak consistency | ✅ Strong consistency |
| Time-consuming | ✅ End-to-end efficient |
| Subjective thresholding | ✅ Objective optimization |
brain-graph-augmentation-template - Graph augmentation methodsmultimodal-brain-connectivity-gnn - Multimodal GNNdrl-gnn-brain-network - Deep RL for brain networksgenerative-brain-dynamics-models - Generative approaches