Computational Complexity Reduction of Neural Networks of Brain Tumor Image Segmentation by Introducing Fermi-Dirac Correction Functions.

Yen-Ling Tai,Chien-Chang Chen,Henry Horng-Shing Lu,Shin-Jhe Huang

doi:10.3390/e23020223

Abstract

Nowadays, deep learning methods with high structural complexity and flexibility inevitably lean on the computational capability of the hardware. A platform with high-performance GPUs and large amounts of memory could support neural networks having large numbers of layers and kernels. However, naively pursuing high-cost hardware would probably drag the technical development of deep learning methods. In the article, we thus establish a new preprocessing method to reduce the computational complexity of the neural networks. Inspired by the band theory of solids in physics, we map the image space into a noninteraction physical system isomorphically and then treat image voxels as particle-like clusters. Then, we reconstruct the Fermi–Dirac distribution to be a correction function for the normalization of the voxel intensity and as a filter of insignificant cluster components. The filtered clusters at the circumstance can delineate the morphological heterogeneity of the image voxels. We used the BraTS 2019 datasets and the dimensional fusion U-net for the algorithmic validation, and the proposed Fermi–Dirac correction function exhibited comparable performance to other employed preprocessing methods. By comparing to the conventional z-score normalization function and the Gamma correction function, the proposed algorithm can save at least 38% of computational time cost under a low-cost hardware architecture. Even though the correction function of global histogram equalization has the lowest computational time among the employed correction functions, the proposed Fermi–Dirac correction function exhibits better capabilities of image augmentation and segmentation.

Highlights

Deep learning methods at present are playing an indispensable role in the field of computer vision
The emergence of fully convolutional neural networks (FCN) has successfully acquired more attention, and the FCN-based methods have further elevated the performance of convolutional neural networks (CNNs) in the field of modern medical image recognition and segmentation [15,18,19,20]
In applications of clinical practice, on the other hand, the medical image datasets are often sparse, so the technical development in these fields is leading by the NN models that are suitable for dealing with small-size datasets

Summary

Introduction

Deep learning methods at present are playing an indispensable role in the field of computer vision. The scope of application covers the demands of computer-guided pathological inspection [1,2,3,4], brain neural circuit mapping and tracking [1,5,6,7,8,9], specific tissue detection in image-based datasets [5,10,11,12,13], and other clinical applications Among these applications, neural network (NN)-based recognition methods that are capable of detecting life-threatening abnormalities from image-based datasets especially attract the attention of both scientific and engineering participants [2,11,14,15,16,17] and gradually replace conventional approaches. Due to the extremely high mortality [1,13], the modality investigations of malignant brain tumors have affected the mainstream techniques of the brain tumor image segmentation [1,30] and the procedures of image-guided surgery

Methods

Results

Conclusion