Raisi-Nafchi Mahsa, Tajmirriahi Mahnoosh, Rabbani Hossein, Amini Zahra
Bioimaging and Biomedical Engineering Department, School of Advanced Technologies in Medicine, Isfahan University of Medical Sciences, Isfahan, 81746734641, Iran.
Bioelectrics and Biomedical Engineering Department, Medical Image and Signal Processing Research Centre, School of Advanced Technologies in Medicine, Isfahan University of Medical Sciences, Isfahan, 81746734641, Iran.
Sci Rep. 2025 Jul 2;15(1):22835. doi: 10.1038/s41598-025-06144-0.
Astrocytomas are among the most prevalent primary brain tumors and are classified into four grades by the World Health Organization. Accurate grading is essential for guiding treatment, as therapeutic strategies depend heavily on tumor grade. This paper presents a new preoperative classification method for astrocytomas, addressing the issue of data scarcity in medical imaging. This work leverages an advanced statistical modeling approach based on stochastic differential equations to analyze post-contrast T1-weighted brain MRI images that require minimal data and offer rapid processing times. In this method, the alpha-stable nature of MRI images is represented by applying a fractional Laplacian filter, and the parameters of the resulting alpha-stable distribution are fed to classifiers to detect the grade of astrocytomas. The method is implemented in both 1D and 2D processing modes, with customized preprocessing for each. Three classification algorithms were evaluated: support vector machine, K-nearest neighbor, and random forest. In the three-class classification task (Grades II-IV), the support vector machine exhibited superior performance, achieving accuracy, sensitivity, and specificity of 98.49%, 98.42%, and 99.23% in 2D mode, and 93.52%, 93.23%, and 96.72% in 1D mode. The results indicate that the proposed framework has the potential to significantly enhance preoperative grading of astrocytomas.
星形细胞瘤是最常见的原发性脑肿瘤之一，世界卫生组织将其分为四个等级。准确分级对于指导治疗至关重要，因为治疗策略在很大程度上取决于肿瘤等级。本文提出了一种新的星形细胞瘤术前分类方法，以解决医学影像中数据稀缺的问题。这项工作利用基于随机微分方程的先进统计建模方法，来分析对比增强后的T1加权脑MRI图像，该方法所需数据最少且处理速度快。在这种方法中，通过应用分数拉普拉斯滤波器来表示MRI图像的α稳定特性，并将所得α稳定分布的参数输入分类器以检测星形细胞瘤的等级。该方法以一维和二维处理模式实现，每种模式都有定制的预处理。评估了三种分类算法：支持向量机、K近邻和随机森林。在三类分类任务（II-IV级）中，支持向量机表现出卓越的性能，在二维模式下准确率、灵敏度和特异性分别达到98.49%、98.42%和99.23%，在一维模式下分别为93.52%、93.23%和96.72%。结果表明，所提出的框架有可能显著提高星形细胞瘤的术前分级。
