Muthanna Journal of Engineering and Technology
Volume (12), Issue (2), Year (30 December 2024), Pages (01-08)
DOI:10.52113/3/eng/mjet/2024-12-02/01-08
Research Article By:
Ahmed Saaudi, Riyadh Mansoor and Salah Alheejawi
Corresponding author E-mail: ahmed.saaudi@mu.edu.iq
ABSTRACT
A brain tumor is a serious disease that requires a talented specialist to differentiate the tumor types (benign or malignant) accurately in the early stages. Artificial intelligent (AI) can participate in facilitating the specialists’ task by performing deep learning algorithms on MRI-based images to achieve an accurate decision. There are many pretrained models that are developed based on deep learning algorithms to tackle brain tumor identification issue. In this work, the weights of three pretrained models: Xception, Inception-resnet50, and VGG 16 are adapted to develop brain-tumor detection systems. The structure of each system is upgraded by adding an input layer and two dense layers of 32, and 16 nodes, respectively, with one output layer to classify the input samples, (MRI of brain tumor). Later, a comprehensive comparison is conducted to evaluate the behavior of each model according to the ability to identify the tumor type (Healthy or malignant). The comparison study reveals the superiority of the VGG16 model in terms of accuracy. Moreover, the structure of the VGG16 model presents less complexity regarding the number of CNN layers and training parameters. Reducing the complexity participates in saving the consuming energy and reducing the execution time. The latter is crucial since it helps specialists to identify the tumor type easily and in, relatively, less time. The main aim of this study is to develop a highly accurate and less complex CNN-based model to recognize the brain tumor type. A model of three CNN-based layers with 719281 trainable parameters is suggested. The proposed model shows 96% accuracy in approximately 8500 seconds. Even though the accuracy of CNN-based model is less than the VGG16 model, the proposed model surpasses the VGG16 in terms of complexity and execution time. Also, the proposed model shows a better performance compared to the Inception-resnet50 v2 and exception models.
Keywords: Deep learning, Transfer learning, AI, Healthcare.