PD-Detection-Models

Multi-modal Parkinson’s Disease detection pipelines featuring models for spiral drawing, tremor, and audio-based analysis.

Part 1: Spiral and wave drawings Model

Models compared:

Model used:

MobileNetV3LargeBinary with modified classifier.

Dataset size:

• 2611: Training
• 653: Validation

Number of trained epochs:

• 20: initial model comparisons
• 50: fine-tuning with early stopping; planned 50.

Metrics:

• Accuracy: 98.81%
• Recall: 99.27%
• precision: 98.30%
• F1-Score: 98.75%

---

Part 2: Tremor Model

Model

• Input → Tremor features + movement type + handedness

• Embeddings → Convert categorical inputs to learned vectors

• Projection → Map all inputs to a shared hidden space

• Feature Attention → Learn which features matter most

• Residual Blocks → Deep feature refinement with stable training

• Classifier → Single output logit (Healthy vs. Parkinson’s)

Generated data (used in pre-training):

• Model used: TVAE (Tabular VAE)
• Generated data accuracy:
  • Column Shapes Score: 92.53%
  • Column Pair Trends Score: 96.8%
  • Overall Score (Average): 94.67%
• Generated: 100k samples
  • 80K: Training
  • 20K: Validation

Metrics:

• Validation Accuracy: 92.27%
• Validation Recall: 92.08%
• Validation precision: 92.5%
• Validation F1-Score: 92.29%

• Validation Accuracy: 90.15%
• Validation Recall: 79.48%
• Validation precision: 76.9%
• Validation F1-Score: 78.17%

---

Part 3: Audio Model (Tabular)

Models compared:

Model used:

DenseNet169.

Generated data (used in pre-training):

• Model used: TVAE (Tabular VAE)
• Generated data accuracy:
  • Column Shapes Score: 87.43%
  • Column Pair Trends Score: 92.2%
  • Overall Score (Average): 89.81%
• Generated: 100k samples
  • 80K: Training
  • 20K: Validation

Real data (used in finetuning):

Note: We used a 20:80 train–test split, allocating less data for training and more for testing, since the model had already been trained on a large amount of generated data and demonstrated strong performance.

• 16: Training
• 65: Validation

Note: The model becomes unstable when fine-tuned for too many epochs, so fine-tuning was stopped early after 7–10 epochs.

Metrics:

• Validation Accuracy: 98.75%
• Validation Recall: 100.00%
• Validation precision: 96.67%
• Validation F1-Score: 98.18%

---

Part 4: Subject's Metadata Model

Models compared:

Model used:

DenseNet169.

Generated data (used in pre-training):

• Model used: TVAE (Tabular VAE)
• Generated data accuracy:
  • Column Shapes Score: 91.69%
  • Column Pair Trends Score: 87.36%
  • Overall Score (Average): 89.53%
• Generated: 100k samples
  • 80K: Training
  • 20K: Validation

Real data (used in finetuning):

• 284: Training
• 71: Validation

Metrics:

• Validation Accuracy: 99.22%
• Validation Recall: 99.06%
• Validation precision: 100.00%
• Validation F1-Score: 99.52%