eduzhai > Applied Sciences > Engineering >

A Computation-Efficient CNN System for High-Quality Brain Tumor Segmentation

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: In this paper, a Convolutional Neural Network (CNN) system is proposed forbrain tumor segmentation. The system consists of three parts, a pre-processingblock to reduce the data volume, an application-specific CNN(ASCNN) to segmenttumor areas precisely, and a refinement block to detect remove false positivepixels. The CNN, designed specifically for the task, has 7 convolution layers,16 channels per layer, requiring only 11716 parameters. The convolutionscombined with max-pooling in the first half of the CNN are performed tolocalize tumor areas. Two convolution modes, namely depthwise convolution andstandard convolution, are performed in parallel in the first 2 layers toextract elementary features efficiently. For a fine classification ofpixel-wise precision in the second half of the CNN, the feature maps aremodulated by adding the individually weighted local feature maps generated inthe first half of the CNN. The performance of the proposed system has beenevaluated by an online platform with dataset of Multimodal Brain Tumor ImageSegmentation Benchmark (BRATS) 2018. Requiring a very low computation volume,the proposed system delivers a high segmentation quality indicated by itsaverage Dice scores of 0.75, 0.88 and 0.76 for enhancing tumor, whole tumor andtumor core, respectively, and also by the median Dice scores of 0.85, 0.92, and0.86. The consistency in system performance has also been measured,demonstrating that the system is able to reproduce almost the same output tothe same input after retraining. The simple structure of the proposed systemfacilitates its implementation in computation restricted environment, and awide range of applications can thus be expected.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...