Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.16418369 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866902184306671616 |
|---|---|
| author | Nianzu, Qu |
| author_facet | Nianzu, Qu |
| contents | <p># Cognitive Embedding Diagnostic Transformer (CEDT)</p> <p></p> <p>## Project Overview</p> <p>This repository implements the **Cognitive Embedding Diagnostic Transformer (CEDT)**, a deep learning architecture designed for **personalized diagnosis of psychiatric disorders** using **neuroimaging data**. It addresses limitations in traditional machine learning by integrating symbolic reasoning, probabilistic inference, and attention-based representation learning.</p> <p>CEDT is accompanied by a novel calibration method, **Disorder-Aware Interpretive Calibration (DAIC)**, which enhances interpretability, uncertainty quantification, and clinical relevance.</p> <p>---</p> <p>## Features</p> <p>- **Multimodal Symptom Embedding** with structured attention layers.<br>- **Latent Diagnostic Space** capturing diagnostic uncertainty.<br>- **Multi-Hypothesis Output** with entropy-aware uncertainty estimation.<br>- **Taxonomic Calibration** using domain knowledge from psychiatric ontologies.<br>- **Graph-based Comorbidity Regularization** for clinically aligned predictions.<br>- ️ Extensive ablation and benchmarking across 4 public datasets.</p> <p>---</p> <p>## Datasets</p> <p>Experiments were conducted on:</p> <p>- **ImageNet** (for transfer learning)<br>- **Caltech-256** (object diversity)<br>- **Oxford 102 Flowers** (fine-grained classification)<br>- **Describable Textures Dataset (DTD)** (texture sensitivity)</p> <p>Note: These datasets are used for architectural benchmarking. The clinical framework is designed for neuroimaging symptom data.</p> <p>---</p> <p>## ️ Architecture</p> <p>- **CEDT Core**: Attention-based encoder-decoder with multi-head attention for symptom-to-diagnosis mapping.<br>- **Tok-KAN Blocks**: Tokenized knowledge-aware attention layers (see *Figure 1, Page 7*).<br>- **Graphical Propagation Layer**: Incorporates co-diagnosis structure via Laplacian smoothing.<br>- **DAIC Module**: Combines Auto-Fusion networks, taxonomic embeddings, and entropy gating for calibration (see *Figure 3, Page 11*).<br>- **Prototype-Centered Refinement**: Aligns output with learned diagnostic prototypes (*Figure 4, Page 13*).</p> <p>---</p> <p>## Training & Evaluation</p> <p>- **Framework**: PyTorch<br>- **Optimizer**: AdamW<br>- **Learning Rate**: `1e-4` (cosine annealing)<br>- **Batch Size**: 64<br>- **Epochs**: 300<br>- **Regularizations**: Label smoothing, dropout, contrastive loss<br>- **Evaluation Metrics**: Accuracy, F1 Score, Recall, AUC</p> <p>---</p> <p>## Results Summary</p> <p>| Dataset | Accuracy | F1 Score | AUC |<br>|--------------------|----------|----------|---------|<br>| ImageNet | 83.90% | 82.95% | 88.62% |<br>| Caltech-256 | 89.47% | 88.74% | 91.55% |<br>| Oxford 102 Flowers | 93.66% | 93.12% | 95.45% |<br>| DTD (Textures) | 78.93% | 78.01% | 81.42% |</p> <p>See *Tables 1 and 2, Pages 16–17* for complete benchmarking.</p> <p>---</p> <p>## ⚙️ Installation</p> <p>```bash<br>git clone https://github.com/your_username/cedt-framework.git<br>cd cedt-framework<br>pip install -r requirements.txt</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_16418369 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | NeuroCEDT Nianzu, Qu <p># Cognitive Embedding Diagnostic Transformer (CEDT)</p> <p></p> <p>## Project Overview</p> <p>This repository implements the **Cognitive Embedding Diagnostic Transformer (CEDT)**, a deep learning architecture designed for **personalized diagnosis of psychiatric disorders** using **neuroimaging data**. It addresses limitations in traditional machine learning by integrating symbolic reasoning, probabilistic inference, and attention-based representation learning.</p> <p>CEDT is accompanied by a novel calibration method, **Disorder-Aware Interpretive Calibration (DAIC)**, which enhances interpretability, uncertainty quantification, and clinical relevance.</p> <p>---</p> <p>## Features</p> <p>- **Multimodal Symptom Embedding** with structured attention layers.<br>- **Latent Diagnostic Space** capturing diagnostic uncertainty.<br>- **Multi-Hypothesis Output** with entropy-aware uncertainty estimation.<br>- **Taxonomic Calibration** using domain knowledge from psychiatric ontologies.<br>- **Graph-based Comorbidity Regularization** for clinically aligned predictions.<br>- ️ Extensive ablation and benchmarking across 4 public datasets.</p> <p>---</p> <p>## Datasets</p> <p>Experiments were conducted on:</p> <p>- **ImageNet** (for transfer learning)<br>- **Caltech-256** (object diversity)<br>- **Oxford 102 Flowers** (fine-grained classification)<br>- **Describable Textures Dataset (DTD)** (texture sensitivity)</p> <p>Note: These datasets are used for architectural benchmarking. The clinical framework is designed for neuroimaging symptom data.</p> <p>---</p> <p>## ️ Architecture</p> <p>- **CEDT Core**: Attention-based encoder-decoder with multi-head attention for symptom-to-diagnosis mapping.<br>- **Tok-KAN Blocks**: Tokenized knowledge-aware attention layers (see *Figure 1, Page 7*).<br>- **Graphical Propagation Layer**: Incorporates co-diagnosis structure via Laplacian smoothing.<br>- **DAIC Module**: Combines Auto-Fusion networks, taxonomic embeddings, and entropy gating for calibration (see *Figure 3, Page 11*).<br>- **Prototype-Centered Refinement**: Aligns output with learned diagnostic prototypes (*Figure 4, Page 13*).</p> <p>---</p> <p>## Training & Evaluation</p> <p>- **Framework**: PyTorch<br>- **Optimizer**: AdamW<br>- **Learning Rate**: `1e-4` (cosine annealing)<br>- **Batch Size**: 64<br>- **Epochs**: 300<br>- **Regularizations**: Label smoothing, dropout, contrastive loss<br>- **Evaluation Metrics**: Accuracy, F1 Score, Recall, AUC</p> <p>---</p> <p>## Results Summary</p> <p>| Dataset | Accuracy | F1 Score | AUC |<br>|--------------------|----------|----------|---------|<br>| ImageNet | 83.90% | 82.95% | 88.62% |<br>| Caltech-256 | 89.47% | 88.74% | 91.55% |<br>| Oxford 102 Flowers | 93.66% | 93.12% | 95.45% |<br>| DTD (Textures) | 78.93% | 78.01% | 81.42% |</p> <p>See *Tables 1 and 2, Pages 16–17* for complete benchmarking.</p> <p>---</p> <p>## ⚙️ Installation</p> <p>```bash<br>git clone https://github.com/your_username/cedt-framework.git<br>cd cedt-framework<br>pip install -r requirements.txt</p> |
| title | NeuroCEDT |
| url | https://doi.org/10.5281/zenodo.16418369 |