An Adaptive Eroded Deep Convolutional neural network for brain image segmentation and classification using Inception ResnetV2. (September 2022)

• Record Type: Journal Article
• Authors: Sunsuhi, G.S.; Albin Jose, S.
• Is Part Of: Biomedical signal processing and control. Volume 78(2022)
• Publication Date: 2022-09
• Subjects: Adaptive Eroded Deep Convolutional neural network -- Binary based Boundary box detection -- Inception ResnetV2 -- Magnetic Resonance Imaging -- Brain tumor; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103863
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Binary based boundary box detection is used in preprocessing which resolves effective step of the BT images depending on boundary pixel values. Adaptive Eroded Deep CNN segmentation is proposed which overcomes existing technique limitations by pixel cluster centroid values of meningioma, glioma and pituitary parts of brain. This research work uses Inception Resnet V2, is proposed to provide classification of tumour of BT images on comparison with other state of art. Abstract: In today's scenario, the main challenging issue in medical field is the tumor detection in human brain. An uncontrolled growth of abnormal nerve tissues contributes to brain tumor. This state of abnormal growth leads to malignant cells which causes a serious issue for its effective treatment. An automated process of brain tumour detection has grabbed attention with improved technological development. This research paper focuses on the effort to segment and identify tumor in human brain. The main steps includes preprocessing, segmentation and classification, where the initial one deals with Anisotropic Diffusion filter followed by Binary based Boundary box detection. The novel procedure of segmentation is done with proposed Adaptive Eroded Deep Convolutional neural network (AEDCNN). It enables to provide distinct segmentation between meningioma, glioma and pituitary brain region. The next step of segmentation is proposed Inception resnetV2, which acts as the novel classification method in brain images. The primary stage of separation is to divide the tumor cell region. The algorithm AEDCNNS determines a degree of spatial membership. The Inception holds different information scales which contributes to input image data. Particularly for classification state mission, we focus on three diseases known as meningioma, glioma and pituitary as benign or malignant. With increase in accuracy, the proposed Inception resnetV2 proves an effective machine learning mechanism for image classification. The accuracy and precision are 97.89%, 93.27% respectively. It proves to be efficient tool for physicians working in medical field.