Multivariate Brain Tumor Detection in 3D-MRI Images Using Optimised Segmentation and Unified Classification Model

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• 3D Magnetic Resonance Imaging (3D-MRI) is crucial for brain tumor diagnosis and treatment planning.
• Existing segmentation techniques struggle with accuracy due to issues in initial contour point extraction and overlapping tissue intensities.
• Accurate classification of brain tumors is hindered by challenges in extracting contextual and symmetric features.

Purpose of the Study:

• To develop an advanced segmentation method for precise brain tumor localization and delineation.
• To propose a novel classification approach for improved detection rates of multivariate brain tumors.
• To minimize segmentation errors and enhance the overall diagnostic accuracy in brain tumor analysis.

Main Methods:

• A Duo-step optimised Pyramidal SegNet with a multiscale contrast convolutional attention module for improved contrast and edge extraction.
• Duo-step darning needle optimization and pyramidal level set segmentation with Sobel edge operator for accurate tumor region extraction.
• A 3D brain Unified Neural Network (NN) employing an adaptive multi-layer deep unified encoder to extract 3D contextual and symmetric features.

Main Results:

• The proposed segmentation method effectively minimizes errors by avoiding overlapped tissue intensity distributions.
• The 3D brain Unified NN demonstrated high detection rates by extracting crucial 3D contextual and symmetric features.
• Evaluated on BraTS2020 and Brain Tumor Detection 2020 datasets, the model achieved high precision, recall, and accuracy.

Conclusions:

• The novel Duo-step optimised Pyramidal SegNet and 3D brain Unified NN significantly outperform existing techniques in brain tumor segmentation and classification.
• The proposed methods offer enhanced precision, recall, and accuracy, improving diagnostic capabilities for brain tumors.
• This research provides a robust framework for more accurate and reliable 3D-MRI based brain tumor analysis.