Brain Tumor Detection Based on a Novel and High-Quality Prediction of the Tumor Pixel Distributions

• DOI: 10.48550/arxiv.2308.07495
• Publication Date: 2023-01-01
• Authors: Yanming Sun; Chunyan Wang
• Keywords: Minimum bounding box; Brain tumor

In this paper, we propose a system to detect brain tumor in 3D MRI scans of Flair modality. It performs 2 functions: (a) predicting gray-level and locational distributions the pixels regions (b) generating mask pixel-wise precision. To facilitate data analysis processing, introduced 2D histogram presentation that comprehends distribution pixel-location object. proposed system, particular histograms, which tumor-related feature get concentrated, are established by exploiting left-right asymmetry structure. A modulation function is generated from input each patient case applied histograms attenuate element irrelevant regions. The prediction pixel done 3 steps, on axial, coronal sagittal slice series, respectively. step, result helps identify/remove tumor-free slices, increasing information density remaining be next step. After 3-step removal, reduced minimum bounding box region. used finalize then transformed into mask, means gray level thresholding low-pass-based morphological operations. final determine critical threshold. has been tested extensively with more than one thousand cases datasets BraTS 2018~21. test results demonstrate predicted have high degree similarity true ones. delivers also very good detection results, comparable those state-of-the-art CNN systems mono-modality inputs, achieved at an extremely low computation cost no need for training.