State-of-the-Art CNN Optimizer for Brain Tumor Segmentation in Magnetic Resonance Images

Abstract

Brain tumors have become a leading cause of death around the globe. The main reason for this epidemic is the difficulty conducting a timely diagnosis of the tumor. Fortunately, magnetic resonance images (MRI) are utilized to diagnose tumors in most cases. The performance of a Convolutional Neural Network (CNN) depends on many factors (i.e., weight initialization, optimization, batches and epochs, learning rate, activation function, loss function, and network topology), data quality, and specific combinations of these model attributes. When we deal with a segmentation or classification problem, utilizing a single optimizer is considered weak testing or validity unless the decision of the selection of an optimizer is backed up by a strong argument. Therefore, optimizer selection processes are considered important to validate the usage of a single optimizer in order to attain these decision problems. In this paper, we provides a comprehensive comparative analysis of popular optimizers of CNN to benchmark the segmentation for improvement. In detail, we perform a comparative analysis of 10 different state-of-the-art gradient descent-based optimizers, namely Adaptive Gradient (Adagrad), Adaptive Delta (AdaDelta), Stochastic Gradient Descent (SGD), Adaptive Momentum (Adam), Cyclic Learning Rate (CLR), Adaptive Max Pooling (Adamax), Root Mean Square Propagation (RMS Prop), Nesterov Adaptive Momentum (Nadam), and Nesterov accelerated gradient (NAG) for CNN. The experiments were performed on the BraTS2015 data set. The Adam optimizer had the best accuracy of 99.2% in enhancing the CNN ability in classification and segmentation.

Keywords: Adam; brain tumor; convolutional neural network; deep learning; gradient descent; optimizer; segmentation.

Figure 1

The 5 tumor types with their survival rates for patients aged between 20–44.

Figure 2

The different tumor types with different shapes in four magnetic resonance images (MRI) sequences: (a) T1 MRI sequence, (b) T2 MRI sequence with tumor type edema, (c) T1C MRI sequence with core tumor, and (d) Search Results Web results Fluid attenuation inversion recovery (FLAIR) sequence showing the ground truth of a tumor.

Figure 3

Proposed architecture.

Figure 4

Proposed model flow chart.

Figure 5

Validation accuracy and loss comparison of all optimizers using our proposed architecture: (a) validation accuracy, (b) training accuracy, (c) validation loss, and (d) training loss.

Figure 6

Automatic segmentation results of HGG (a) and LGG (b) cases. Red: edema, blue: non-enhancing tumor. From left to right: (i) original image, (ii) ground truth, and (iii) demonstration of automatic segmentation results from the proposed method. (a) The segmentation results of HGG cases compared to their ground truth and (b) the segmentation results of LGG cases compared to their ground truth.

Figure 7

Accuracy rate and error rate comparison of all optimizers using our proposed architecture: (a) accuracy rate and (b) error rate.