Inspiration
This project was inspired by my research on schizophrenia diagnosis and treatment, which highlighted the need for improved AI-based diagnostic models. As a first step toward applying my findings, I created this proof of concept to demonstrate how analysing functional and structural brain imaging data using ML models can enhance diagnostic accuracy and support personalised psychiatric care.
What it does
The Schizophrenia Diagnostic Tool is a web-based application that accepts structural and functional brain imaging data of different patients and processes and analyses this data using a logistic regression model. The tool classifies patients into two categories: "Schizophrenia" and "Control". The web application displays diagnoses for each patient in the input dataset, along with their respective Patient ID and the confidence level of the diagnosis.
How I built it
Logistic regression model:
- Sigmoid activation for probability transformation
- 0.5 threshold for classification
- Cross-entropy loss function
- Gradient descent optimization
- L1 and L2 regularization
Web Interface:
- Flask for the backend.
- HTML/CSS for the front end.
Challenges
Finding an appropriate dataset with comprehensive functional and structural MRI-derived features was challenging. Publicly available datasets often lacked sufficient labelled data or required significant preprocessing to make them usable for machine learning tasks.
Accomplishments
- Successfully implemented a custom machine learning model tailored for this application.
- Developed a clean and responsive web-based interface that makes the tool accessible and user-friendly for non-technical users.
What we learned
- The importance of robust data preprocessing when working with real-world datasets, particularly when handling missing or noisy data.
- How to design and implement a custom machine learning model from scratch, including optimization techniques like gradient descent and regularization.
- Best practices for building web applications integrating machine learning models to provide real-time predictions.
What's next for Schizophrenia Diagnostic Tool
The next steps involve refining the model using larger, more diverse datasets to improve its robustness. Incorporating insights from my research, I aim to make the tool multimodal, integrating data from multiple sources like genetics and clinical assessments. I plan to leverage explainable AI techniques to enhance transparency and trust. Ultimately, this tool could evolve into a clinical decision-support system, aiding professionals in delivering personalised psychiatric care.
Log in or sign up for Devpost to join the conversation.