Learn about Magnetic Resonance Imaging (MRI) and Diffusion-Weighted Imaging (DWI) - and how this research project uses brain MRI images
This experimental student research project explores how deep learning models respond to single brain MRI images. They often fail on real uploaded scans and are not clinical tools. The models:
Process one 2D MRI slice and flag regions the model considers possible tumors, with model confidence scores
Show boxes or pixel outlines around flagged regions; classification models also predict one of three tumor types
FOR EDUCATION AND RESEARCH ONLY - NOT FOR CLINICAL DIAGNOSIS
MRI is a non-invasive medical imaging technique that uses strong magnetic fields and radio waves to create detailed images of organs and tissues inside the body.
Uses magnetic fields to align hydrogen atoms in your body, then radio waves to create signals that are converted into detailed images
No ionizing radiation like X-rays or CT scans, making it safe for repeated use and detailed studies
Provides excellent contrast between different soft tissues, perfect for brain and tumor imaging
DWI is a specialized MRI technique that measures the movement of water molecules in tissues. This movement pattern can reveal important information about tissue structure and health.
DWI is commonly used for:
Researchers study whether deep learning can help identify patterns in medical images. This project tests three models trained on brain MRI images, but strong development results have not carried over to real uploads.
Choose a model output to explore; the classification models only know three tumor types:
The classification results include an interactive symptom checklist that compares the model's predicted type with reported symptoms. This educational tool calculates a "Combined Consistency Score" based on model confidence and checklist responses; it is not clinically validated.
Headaches, seizures, cognitive changes, motor weakness, speech difficulties
Headaches, vision changes, hearing loss, limb weakness, memory problems
Vision changes, hormonal symptoms, fatigue, unexplained weight changes
The Combined Consistency Score weighs AI confidence (55%) and symptom checklist responses (45%) as illustrative values for this research demo. They are not clinically validated and do not provide a diagnosis.
One motivation for medical imaging AI research is to study whether tools might someday help with review workloads. This project has not shown that its models save hospital time or improve patient care.
Scanners, MRI sequences and hospital protocols change how images look. A model trained on one collection may flag the wrong regions or miss tumors on scans from another hospital.
Testing on images from other hospitals and scanners is needed before anyone can judge whether a model helps in a clinical workflow. The models here have not passed that test.
These models see only one 2D slice. Radiologists read full 3D studies with several sequences and clinical history. Understanding how data changes across hospitals, and checking results on external scans, must come before any claim that an AI tool could make care more efficient.