Tumor Segmentation in Contrast-Enhanced Magnetic Resonance Imaging for Nasopharyngeal Carcinoma: Deep Learning with Convolutional Neural Network

Abstract

Objectives: To evaluate the application of a deep learning architecture, based on the convolutional neural network (CNN) technique, to perform automatic tumor segmentation of magnetic resonance imaging (MRI) for nasopharyngeal carcinoma (NPC).

Materials and methods: In this prospective study, 87 MRI containing tumor regions were acquired from newly diagnosed NPC patients. These 87 MRI were augmented to >60,000 images. The proposed CNN network is composed of two phases: feature representation and scores map reconstruction. We designed a stepwise scheme to train our CNN network. To evaluate the performance of our method, we used case-by-case leave-one-out cross-validation (LOOCV). The ground truth of tumor contouring was acquired by the consensus of two experienced radiologists.

Results: The mean values of dice similarity coefficient, percent match, and their corresponding ratio with our method were 0.89±0.05, 0.90±0.04, and 0.84±0.06, respectively, all of which were better than reported values in the similar studies.

Conclusions: We successfully established a segmentation method for NPC based on deep learning in contrast-enhanced magnetic resonance imaging. Further clinical trials with dedicated algorithms are warranted.

Figure 1

The schematic diagram of the proposed convolutional neural network (CNN) structure. The proposed CNN network includes two phases of feature representation and scores map reconstruction. The feature representation phase consists of 2 Pool-Conv-ReLu blocks and 3 Con-ReLu blocks, while the scores map reconstruction phase consists of 2 deconv-concat-conv blocks. The output of each layer is a three-dimensional matrix with size of h×w×d, where h and w are the length and width of the scores map, respectively, and d is the feature dimension. a×a indicates the matrix size of the convolution kernels. Conv: convolution, Relu: rectified linear units, Pool: pooling, Deconv: deconvolution.

Figure 2

NPC tumor segmentation with high accuracy using the current deep learning method with convolutional neural network. (a) The original image. (b) Ground truth (white line). (c) Segmentation from our deep learning method result (white line) with the dice similarity coefficient = 0.941, corresponding ratio = 0.915, and percent match = 0.950.

Figure 3

NPC tumor segmentation with less accuracy using the current deep learning method with convolutional neural network. (a) The original image. (b) Ground truth (white line). (c) Segmentation from our deep learning method result (white line) with the dice similarity coefficient = 0.797, corresponding ratio = 0.731, and percent match = 0.937.

Figure 4

A feature map fusion. (a) Scores map acquired in the reconstruction phase and (b) scores map reconstructed from the fused feature maps, indicating that score map reconstruction using fused maps is better than that during reconstruction phase.