NeuroAid

Inspiration

The inspiration for NeuroAid comes from the growing global impact of Alzheimer’s disease and the need for early, accessible detection tools. Alzheimer’s often goes undiagnosed until significant cognitive decline occurs, limiting treatment effectiveness. Advances in medical imaging and deep learning inspired us to explore how AI—specifically CNNs—can assist in identifying subtle brain changes from MRI scans and support earlier clinical awareness.

What It Does

NeuroAid uses a Convolutional Neural Network to analyze grayscale MRI brain images and classify them as either Healthy or Alzheimer’s. The system automatically processes MRI data, extracts spatial features related to brain structure, and predicts disease presence with measurable confidence. It provides performance metrics, visualizations, and prediction outputs to support clinical research and educational use.

How We Built It

We built NeuroAid using PyTorch for deep learning and Python-based data science tools for preprocessing and evaluation. MRI images were extracted from a Parquet dataset, converted into normalized grayscale arrays, and fed into a custom CNN architecture. The model was trained using weighted loss to handle class imbalance and evaluated using accuracy, classification reports, and visual inspection of predictions. GPU acceleration was utilized when available to optimize training.

Challenges We Ran Into

One major challenge was handling class imbalance, as healthy and Alzheimer’s samples were not evenly distributed. Another challenge was ensuring consistent preprocessing when converting raw image bytes into usable tensors. We also faced limitations in model confidence and validation accuracy plateauing, indicating the need for deeper architectures or improved feature extraction.

Accomplishments That We’re Proud Of

We successfully developed an end-to-end deep learning pipeline capable of reading Parquet-based MRI data, training a CNN model, and producing reliable predictions. Achieving around 85% validation accuracy and correctly classifying all randomly selected test samples demonstrates the model’s effectiveness. The project also includes strong visualization and evaluation components, making results interpretable and transparent.

What We Learned

NeuroAid was a highly motivating project that turned theory into practical expertise in medical image preprocessing, CNN design, and handling imbalanced datasets. We learned to prioritize model reliability beyond accuracy, understanding how preprocessing, loss weighting, and evaluation metrics impact performance. Most importantly, the project emphasized the responsibility of using AI in healthcare ethically and transparently, inspiring us to build AI solutions that are both technically strong and socially meaningful.

What’s Next for NeuroAid

Future enhancements will focus on making NeuroAid even more powerful and clinically useful. Plans include expanding the model to multi-class Alzheimer’s stage prediction to provide more detailed insights into disease progression. We aim to incorporate data augmentation, advanced preprocessing, and deeper CNN or transfer learning architectures to improve prediction accuracy and robustness.

We also plan to implement confidence calibration and explainability tools like Grad-CAM, enabling clinicians and researchers to understand model decisions and trust predictions. In the long term, NeuroAid will be tested and validated for real-world clinical use, supporting early detection, monitoring, and decision-making, ultimately helping patients and healthcare providers take informed actions.