[ONL304] - A Submission for the IBM Z "Technology for Good" Datathon
FOCUS (Forwarding Outcomes with Computational Understanding and Support) is a proof-of-concept for an AI-driven platform that provides early, objective, and explainable ADHD risk detection using raw EEG data. It is architected for the secure and scalable IBM Z enterprise environment, aligning with the "Technology for Good" theme by addressing critical challenges in mental health diagnostics.
Attention-Deficit/Hyperactivity Disorder (ADHD) affects millions, yet its diagnosis relies heavily on subjective interviews and behavioral checklists. This process can be slow, inconsistent, and lead to delayed intervention, negatively impacting academic performance, social development, and long-term mental health.
Our project addresses this by leveraging AI to analyze objective neurological data, providing clinicians with a powerful tool for faster and more reliable screening.
FOCUS is an end-to-end pipeline that transforms raw brainwave data into actionable insights via a simple, interactive dashboard.
The user begins by uploading a sample EEG data file. The interface is clean and intuitive, designed for clinical use.
Upon clicking "Analyze," the AI model provides a clear risk assessment with a quantifiable confidence score. Our model achieved 99.91% accuracy on the test data.
Positive ADHD Detection:
Neurotypical (Control) Detection:
Alongside the diagnosis, the system provides actionable, personalized recommendations based on the AI's findings, helping to guide the next steps for the user or clinician.
A key innovation of FOCUS is its explainability. Our AI doesn't just deliver a prediction; it highlights the key brain regions that most influenced its decision. This is critical for building clinical trust and providing deeper diagnostic insights. Our analysis identified the Frontal Lobe channels (F3, F4, Fz) as key indicators, a finding that aligns with clinical research on executive function in ADHD.
- Clone this repository:
git clone https://github.com/realyashagarwal/FOCUS.git
- Set up the environment:
cd FOCUS pip install -r requirements.txt - Launch the Dashboard:
A new tab will open in your browser. You can use the provided
streamlit run app.py
test_sample_ADHD.csvortest_sample_Control.csvfiles to verify the predictions.
While this demo runs locally, it is architected for the power and security of the IBM Z platform.
- Secure Data Management: In production, sensitive patient EEG data would be stored in a Db2 for z/OS database, protected by the mainframe's pervasive encryption.
- Real-Time, Low-Latency Inference: Our deployment path leverages the ONNX format and the IBM Deep Learning Compiler to create an inference engine hyper-optimized for the Telum AI accelerator.
- Scalability for Nationwide Impact: Hosted on Watsonx on Z, our solution can reliably serve thousands of users simultaneously, making large-scale mental health screening a reality.
- AI/ML Frameworks: TensorFlow (Keras), Scikit-learn
- Data Processing: Pandas, NumPy
- Interactive Dashboard: Streamlit
- Target Enterprise Platform: IBM Z (LinuxONE) with Db2 for z/OS, ONNX, and Watsonx on Z
app.py: The Python script for our interactive Streamlit dashboard.ADHD_Detection_IBM_Z_PoC.ipynb: The main Jupyter Notebook showing the complete data processing, model training, and explainability analysis performed on the IBM Z environment.requirements.txt: A list of all necessary Python libraries.test_sample_*.csv: Small, verified data samples for testing the dashboard..h5&.pklfiles: The saved, pre-trained AI model and preprocessing objects.assets/: A folder containing all the screenshots for this README.