Brain tumor segmentation based on optimized convolutional neural network and improved chimp optimization algorithm

作     者：Ranjbarzadeh, Ramin Zarbakhsh, Payam Caputo, Annalina Tirkolaee, Erfan Babaee Bendechache, Malika 

作者机构：Dublin City Univ Fac Engn & Comp Sch Comp Dublin Ireland Cyprus Int Univ Elect & Elect Engn Dept Via Mersin 10 Nicosia Northern Cyprus Turkiye Istinye Univ Dept Ind Engn Istanbul Turkiye Yuan Ze Univ Dept Ind Engn & Management Taoyuan Taiwan Lebanese Amer Univ Dept Ind & Mech Engn Byblos Lebanon Univ Galway Lero & ADAPT Res Ctr Sch Comp Sci Galway Ireland 

出 版 物：《COMPUTERS IN BIOLOGY AND MEDICINE》 (生物学与医学中的计算机)

年 卷 期：2024年第168卷

页      面：107723-107723页

核心收录：

学科分类：0831[工学-生物医学工程（可授工学、理学、医学学位）] 0710[理学-生物学] 07[理学] 09[农学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Science Foundation Ireland [18/CRT/6183  13/RC/2106_P2  13/RC/2094_P2] 

主　　题：Brain tumor Deep learning Improved chimp optimization algorithm Convolutional neural network Feature selection 

摘      要：Reliable and accurate brain tumor segmentation is a challenging task even with the appropriate acquisition of brain images. Tumor grading and segmentation utilizing Magnetic Resonance Imaging (MRI) are necessary steps for correct diagnosis and treatment planning. There are different MRI sequence images (T1, Flair, T1ce, T2, etc.) for identifying different parts of the tumor. Due to the diversity in the illumination of each brain imaging modality, different information and details can be obtained from each input modality. Therefore, by using various MRI modalities, the diagnosis system is capable of finding more unique details that lead to a better segmentation result, especially in fuzzy borders. In this study, to achieve an automatic and robust brain tumor segmentation framework using four MRI sequence images, an optimized Convolutional Neural Network (CNN) is proposed. All weight and bias values of the CNN model are adjusted using an Improved Chimp Optimization Algorithm (IChOA). In the first step, all four input images are normalized to find some potential areas of the existing tumor. Next, by employing the IChOA, the best features are selected using a Support Vector Machine (SVM) classifier. Finally, the best-extracted features are fed to the optimized CNN model to classify each object for brain tumor segmentation. Accordingly, the proposed IChOA is utilized for feature selection and optimizing Hyperparameters in the CNN model. The experimental outcomes conducted on the BRATS 2018 dataset demonstrate superior performance (Precision of 97.41 %, Recall of 95.78 %, and Dice Score of 97.04 %) compared to the existing frameworks.