A multiscale morphological segmentation and classification of brain tumor using supervised learning algorithm

• Published in: AIP conference proceedings Vol. 2966; no. 1
• Main Authors: Krishnan, Pavihaa Lakshmi Babu Muthu, Sampath, Vidhya
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Melville American Institute of Physics 26.03.2024
• Subjects: Algorithms; Brain; Brain cancer; Datasets; Discrete cosine transform; Fourier transforms; Image segmentation; K-nearest neighbors algorithm; Lesions; Machine learning; Magnetic resonance imaging; Medical imaging; Morphology; Noise reduction; Supervised learning; Support vector machines; Tumors
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0189822

A Brain tumor segmentation and classification method is developed using K-nearest neighbor (KNN), a supervised machine learning algorithm for classification to classify the tumor types. The input dataset is filtered using bilateral filtering by preserving edges and reducing noise. The gray level images of Magnetic resonance imaging (MRI) brain tumor are segmented using K-means Multiscale morphology and the Discrete Cosine Transform (DCT), Pigmentation, orientation, Discrete Fourier Transform (DFT), Lesion margin, lesion intensity, lesion variation features of segmented images are analyzed. Based on the features the input dataset is classified by using KNN algorithm. As a result, the MRI brain tumor is classified into 4 types Glioma tumor, Pituitary tumor, Meningioma tumor, no tumor. The proposed KNN algorithm is compared with support vector machine algorithm and this work eventually proved that the performance of KNN is very good than Support vector machines (SVM) in classifying tumor types. The performance analysis calculated using accuracy, precision, recall, g-mean, sensitivity and specificity. MATLAB R2019b is used as the simulation tool for the analysis.