Journal of Engineering Research
Abstract
Brain tumors are a leading cause of death worldwide. With the advancements in medicine and deep learning technologies, the dependency on manual classification-based diagnosis drives down owing to their inaccurate diagnosis and prognosis. Accordingly, the proposed model provides an accurate multi-class classification model for brain tumor using the convolution neural network (CNN) as a backbone. Our novel model is based on concatenating the extracted features from the proposed three branches of CNN, where each branch is fed by the output of different transform domains of the original magnetic resonance image (MRI). These transformations include Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), and the time-domain of the original image. Then, the CNN is employed followed by a concatenation layer, flatten laver, and dense layer, before using the SoftMax layer. The proposed model was applied to the Figshare dataset of brain tumor which consists of three classes pituitary, glioma, and meningioma. The results proved the advantage of the proposed system which achieved a high mean performance over 5-fold cross-validation with 98.89% accuracy, 98.78% F1-score, 98.74% precision, 98.82% recall, and 99.44% specificity. The comparative study with well-known models, as well as the pre-trained CNN models, established the potential of the proposed model. This novel approach has the potential to significantly improve brain tumor classification accuracy. It enables a more comprehensive and objective analysis of brain tumors, leading to improved treatment decisions and better patient care.
Recommended Citation
S. Abd El-Wahab,, Basant
(2023)
"A Novel Convolutional Neural Network Based on Combined Features from Different Transformations for Brain Tumor Diagnosis,"
Journal of Engineering Research: Vol. 7:
Iss.
2, Article 50.
Available at:
https://digitalcommons.aaru.edu.jo/erjeng/vol7/iss2/50