Neuro-Regenerative Multimodal Detection

Project Overview

This project focuses on the detection of Alzheimer's disease using the OASIS-1 dataset. It employs a multimodal approach, integrating Exploratory Data Analysis (EDA), Convolutional Neural Networks (CNN), and anomaly detection techniques to analyse MRI data.

Dataset

The dataset used in this project is OASIS-1 (Open Access Series of Imaging Studies).

• Source: OASIS-1 Dataset
• Description: A cross-sectional MRI data set in young, middle-aged, nondemented, and demented older adults.

Project Structure

The project is organized into sequential steps, each implemented in a Jupyter Notebook:

1. EDA_STEP_1.ipynb: Exploratory Data Analysis. Visualizing distributions, correlations, and initial data insights.
2. DATAprep_STEP_2.ipynb: Data preprocessing. Cleaning, normalization, and preparation of MRI slices for modeling.
3. Baseline_STEP_3.ipynb: Establishing baseline models for performance comparison.
4. CNN_STEP_4.ipynb: Implementation of Convolutional Neural Networks for classification/detection.
5. Testing_STEP_5.ipynb: Evaluation of the trained models on test sets.
6. Anomaly_STEP_6.ipynb: Anomaly detection to identify outliers or unusual patterns in the imaging data.
7. GRAD_STEP_7.ipynb: Gradient-based visualization (e.g., Grad-CAM) to explain model decisions and highlight regions of interest in MRI scans.

Directory Layout

• OUTPUTS/: Contains intermediate outputs, preprocessed data, and model artifacts.
• PROCESSED/: Directory for processed image files.
• eda_outputs/: Generated figures and statistics from the EDA step.
• explainability_outputs/: Visualizations from the explainability step.
• gradcam_outputs/: Grad-CAM heatmaps.
• logs/: Training logs.
• models/: Saved trained models (.h5 files like cnn_v1_best.h5, cnn_v1_final.h5).
• results/: Evaluation results and metrics.

Requirements

To run this project, you will need the following libraries:

• Python 3.x
• TensorFlow / Keras
• NumPy
• Pandas
• Matplotlib
• Seaborn
• Tune (if used for hyperparameter tuning)
• OpenCV (cv2)
• Scikit-learn
• Nibabel (for MRI file handling)

Usage

1. Clone the repository.
2. Download the OASIS-1 dataset from the link provided above.
3. Ensure the data paths in the notebooks correspond to your local setup.
4. Run the notebooks in the numerical order (STEP_1 to STEP_7) to reproduce the pipeline.

Author

Anshul Yadav