Confidence Region Identification and Contour Detection in MRI Image

Area of Science:

• Medical Imaging
• Radiology
• Computational Biology

Background:

• Tumor region extraction (RE) in MRI is challenging due to intensity variations causing poor visibility and mixing of tumor tissues with normal brain tissues.
• Accurate identification of tumor boundaries is crucial for effective diagnosis and treatment planning.
• Existing methods struggle with the subtle intensity differences that define tumor growth regions.

Purpose of the Study:

• To develop a robust method for brain tumor region extraction in MRI images.
• To enhance the accuracy of tumor boundary identification using a confidence score.
• To improve the visualization and analysis of tumor growing regions.

Main Methods:

• A novel region extraction method utilizing confidence scores derived from intensity patterns in MRI images.
• Analysis of intensity patterns, scaling, and identification of the largest connected components (blobs).
• Integration of contour detection (CD) with region scale fitting to delineate tumor boundaries, followed by conversion to confidence regions (CR) using confidence intervals (CI).

Main Results:

• The proposed method achieved high evaluation metrics: 97% Dice Over Index (DOI) and 94% Jacquard Index (JI).
• Mean Squared Error (MSE) was 1.24, and Peak Signal-to-Noise Ratio (PSNR) was 17.45, outperforming benchmark values.
• Accurate classification of mean and deviating pixel values (intensities) effectively highlighted tumor regions.

Conclusions:

• The developed method accurately extracts tumor regions and identifies boundaries with high confidence.
• The confidence score and region approach effectively addresses the challenge of varying MRI image intensities.
• This technique offers a significant advancement in the precise localization and analysis of brain tumors in MRI data.