Ultrasound Scatterer Density Classification Using Convolutional Neural Networks by Exploiting Patch Statistics

TL;DR

This work proposes a convolutional neural network architecture for QUS, and trains it using simulation data, and improves the network performance by utilizing patch statistics as additional input channels, and demonstrates the potential of CNNs in classifying scatterer density in ultrasound images.

Abstract

Quantitative ultrasound (QUS) can reveal crucial information on tissue properties such as scatterer density. If the scatterer density per resolution cell is above or below 10, the tissue is considered as fully developed speckle (FDS) or low-density scatterers (LDS), respectively. Conventionally, the scatterer density has been classified using estimated statistical parameters of the amplitude of backscattered echoes. However, if the patch size is small, the estimation is not accurate. These parameters are also highly dependent on imaging settings. In this paper, we propose a convolutional neural network (CNN) architecture for QUS, and train it using simulation data. We further improve the network performance by utilizing patch statistics as additional input channels. We evaluate the network using simulation data, experimental phantoms and in vivo data. We also compare our proposed network with different classic and deep learning models, and demonstrate its superior performance in classification of tissues with different scatterer density values. The results also show that the proposed network is able to work with different imaging parameters with no need for a reference phantom. This work demonstrates the potential of CNNs in classifying scatterer density in ultrasound images.

Figures (7)

• Figure 1: The distribution of the patch statistics for FDS and LDS in simulated training data. The patch size is small enough such that FDS and LDS classes overlap.
• Figure 2: Proposed architecture for different networks.
• Figure 3: ROC curves of different models on simulation test data.
• Figure 4: ROC curves of experimental phantoms. There is a considerable difference between single channel CNNs with/without utilizing patch statistics. CNN 5 and CNN 6 which use $A\times log(A)$ and $A$ as input channels perform well for experimental data. Fine-tuning further improve CNN 6 by removing domain shift.
• Figure 5: The results of MLP, CNN 5, CNN 6 and CNN 6 + ft models on the experimental phantoms. The color code represents the predicted output of the networks, from 0 (LDS) to 1 (FDS).