A Deep Learning Model Based on Concatenation Approach for Classification of MRI Images

Abstract

The heterogeneity of the tumor cells makes it difficult for radiologists to classify brain tumors, a potentially fatal condition. As an assistive technology, computer-aided diagnosis-based systems have recently promised to use magnetic resonance imaging (MRI) to diagnose brain tumors. Typically, features are extracted from bottom layers in recent applications of pre-trained models, which are distinct from natural and medical images. This study proposes a method for the early diagnosis of brain tumors using multi-level feature extraction and concatenation to solve this issue. This model is valid thanks to the two pretrained deep learning models Inception-v3 and DensNet201. Two distinct scenarios for the detection and classification of brain tumors were evaluated with the assistance of these two models. For brain tumor classification, the features from various Inception modules were first extracted from the pre-trained Inception-v3 model and combined. The softmax classifier was then used to classify the brain tumor using these features. Second, features were extracted from various DensNet blocks using pre-trained DensNet201. The softmax classifier was then used to classify the brain tumor after these features were concatenated. A publicly accessible three-class brain tumor dataset was used to evaluate both scenarios. With Inception-v3 and DensNet201 on testing samples, the proposed method produced 99.34% and 99.51% testing accuracies, respectively, and performed best in the detection of brain tumors. For brain tumor classification, the results indicated that the proposed method based on features concatenation and using pre-trained models performed better than the current state-of-the-art deep learning and machine learning methods.