A hybrid deep CNN-Cov-19-Res-Net Transfer learning architype for an enhanced Brain tumor Detection and Classification scheme in medical image processing

• Published in: Biomedical signal processing and control 2022-07, Vol.76, p.103631, Article 103631
• Main Authors: Ananda Kumar, K.S., Prasad, A.Y., Metan, J.
• Format: Article
• Language: English
• Subjects: Brain tumor detection and classification; Covid 19 optimization algorithm; Deep convolutional neural network; ResNet 152 transfer learning
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2022.103631

•In this work, DCNN with ResNet 152 TL model using CNN to classify the brain images.•Input images are pre-processed by Otsu binarization method.•Images are classified using the Hyb-DCNN-ResNet 152 TL.•Hyb-DCNN-ResNet 152 TL weight parameters are tuned using CoV-19 OA.•The simulation process is executed in the MATLAB platform. The major intention of this work is to detect the Brain tumor with accuracy by reducing error rate and computational complexity. Therefore, in this manuscript, a Deep Convolutional Neural Network with Nature-inspired Res net 152 Transfer Learning model using CNN and Transfer learning tactics is proposed to detect and classify the brain images. Here, the images are pre-processed to remove the noises, also enhance the quality of the images by using Otsu binarization method. The image features, like contrast, Energy, Correlation, Homogeneity, Entropy are extracted with the help of Gray-Level Co-Occurrence Matrix methods. Then, the images are classified using the hybrid Deep Convolutional Neural Network with Nature-inspired Res Net 152 Transfer Learning (Hyb-DCNN-ResNet 152 TL), in which the batch normalization layer of the Deep CNN is removed and added with ResNet 152 layer. Here, hybrid Deep Convolutional Neural Network with Nature-inspired Res Net 152 Transfer Learning classifies as normal, benign and malignant. Then the Hyb-DCNN-ResNet 152 TL weight parameters are tuned using Covid-19 optimization algorithm (CoV-19 OA). The simulation process is executed in the MATLAB platform. The proposed method attains higher accuracy of 99.57%, 97.28%, 94.31%, 95.48%, 96.38%, 98.41% and 96.34%, lower Error rate of 0.012(s), 0.014(s), 0.011(s), 1.052(S), 0.013(S), 0.016(S) and 0.015(s) compared with existing methods, like BTC-Deep CNN-Dolphin-SCA, BTC-Deep CNN-WHHO, BTC-AFDNN-FLA, BTC-MLPNN-IWOA, BTC-ANN-PSO, BTC-RF-WSO and BTC-WRF-ACO.