Before a tumor can be measured, the image has to be made clean enough to trust.
MRI scans are noisy, and noise is the enemy of segmentation: a tumor boundary is only as reliable as the image it is drawn on. This project — my current VNU-HCM research grant (C2025-18-11, 2025–2027) — pairs principled denoising with a fast clustering step to pull tumor regions out of brain MRIs.
The pipeline
The enhancement stage uses an LMMSE estimator together with thresholding built on the stationary wavelet transform and empirical mode decomposition — suppressing noise while preserving the edges that matter clinically. A Fast C-Means clustering stage then segments the cleaned image, separating tumor tissue from healthy structure quickly enough to be practical.
Why it matters
Cleaner segmentation means more reliable tumor measurements, which feed directly into diagnosis and treatment planning. The work was published in IJACSA (Vol. 16, No. 5, 2025) and anchors my ongoing direction in biomedical image processing.
