High precision brain tumor classification model based on deep transfer learning and stacking concepts

Abstract

In this article, we proposed an intelligent clinical decision support system for the detection and classification of brain tumor from risk of malignancy index (RMI) images. To overcome the lack of labeled training data needed to train convolutional neural networks, we have used a deep transfer learning and stacking concepts. For this, we choosed seven convolutional neural networks (CNN) architectures already pre-trained on an ImageNet dataset that we precisely fit on magnetic resonance imaging (MRI) of brain tumors collected from the brain tumor segmentation (BraTS) 19 database. To improve the accuracy of our global model, we only predict as output the prediction that obtained the maximum score among the predictions of the seven pre-trained CNNs. We used a 10-way cross-validation approach to assess the performance of our main 2-class model: low-grade glioma (LGG) and high-grade glioma (HGG) brain tumors. A comparison of the results of our proposed model with those published in the literature, shows that our proposed model is more efficient than those published with an average test precision of 98.67%, an average f1 score of 98.62%, a test precision average of 98.06% and an average test sensitivity of 98.33%.

Convolutional Neural Network Techniques for Brain Tumor Classification (from 2015 to 2022): Review, Challenges, and Future Perspectives

“…Many attempts have been made to investigate the value of transfer learning techniques for brain tumor classification [ 39 , 45 , 50 , 102 , 104 , 108 , 116 , 121 ]. Deepak and Ameer [ 39 ] used the GoogLeNet with the transfer learning technique to differentiate between glioma, MEN, and PT from the dataset provided by Cheng [ 55 ].…”

“…The authors in [ 104 ] used InceptionResNetV2, DenseNet121, MobileNet, InceptionV3, Xception, VGG16, and VGG19, which have already been pre-trained on the ImageNet dataset, to classify HGG and LGG brain images. The MR images used in this research were collected from the BraTS 2019 database, which contains 285 patients (210 HGG, 75 LGG).…”

A review on detecting brain tumors using deep learning and magnetic resonance images

“…The architecture of the DenseNet is composed of four blocks, but each block has a different number of layers. For instance, DenseNet-121 has (6,12,24,16) layers, DenseNet-169 has (6,12,32,32) layers, and DenseNet-201 has [6,12,48,32] layers [60]. Noreen et al [61] presented a model composed of Inception v3 and DenseNet201.…”

“…Due to imprecise findings in detecting brain tumors with a significant variation in decision-making, numerous people were passed away in recent years [12]. Thus, several works have been released [62].…”

“…Support vector machine (SVM) is commonly employed among various machine learning algorithms due to its accuracy and ease of implementation. Recently, machine learning algorithms have been combined with deep learning models to achieve better detecting and diagnosing accuracy [11], [12].…”