Generative AI Trained for Semantic Segmentation of Brain CT and MRIs
D3: Test Plan
Ideal Scenario
Section I: Ideal Testing Scope with Unlimited Resources
Extensive Training and Model Accuracy Testing:
Primary Objective: With unrestricted time and resources, the core focus will be on diligently training the GAN model to ensure high accuracy in segmenting brain MRI scans.
Key Aspects:
Rigorous training cycles to fine-tune the model for precise segmentation of cerebral spinal fluid, brain matter, and other categories.
Implementing comprehensive accuracy tests to evaluate the segmentation results against benchmarks and expert-verified data.
Exploring and Testing Diverse Architectures:
Research and Optimization:
Given the project’s dual nature as a research initiative and an optimization challenge, a significant portion of our efforts would be channeled into exploring and testing a wide array of model architectures.
Testing Approaches:
Conducting parallel tests across different GAN architectures to determine the optimal configuration for accuracy and performance.
Section II: Practical Testing Plan
Unit Testing:
Ensuring proper uploading and processing of NIFTI files.
Verifying each step from preprocessing to running through the model.
Integration and System Testing:
Testing the upload functionality on the web app.
Verifying the processing and model output visualization on the frontend.
Tools Used:
Command-line interface, React for frontend (for class purposes)
Google Colab, Python, TensorFlow.
End User Types:
Hospital staff for processing MRI scans.
Researchers and students for segmenting and studying MRI scans.
Performance and Reliability Testing:
Ensuring the model processes images similar but not identical to the training data.
Acceptance Testing:
Producing reasonable segmented results with minimal unrealistic areas.
Achieving segmentation close to manually segmented images.