Investigation of a Novel Deep Learning-Based Computed Tomography Perfusion Mapping Framework for Functional Lung Avoidance Radiotherapy

Abstract

Functional lung avoidance radiation therapy aims to minimize dose delivery to the normal lung tissue while favoring dose deposition in the defective lung tissue based on the regional function information. However, the clinical acquisition of pulmonary functional images is resource-demanding, inconvenient, and technically challenging. This study aims to investigate the deep learning-based lung functional image synthesis from the CT domain. Forty-two pulmonary macro-aggregated albumin SPECT/CT perfusion scans were retrospectively collected from the hospital. A deep learning-based framework (including image preparation, image processing, and proposed convolutional neural network) was adopted to extract features from 3D CT images and synthesize perfusion as estimations of regional lung function. Ablation experiments were performed to assess the effects of each framework component by removing each element of the framework and analyzing the testing performances. Major results showed that the removal of the CT contrast enhancement component in the image processing resulted in the largest drop in framework performance, compared to the optimal performance (~12%). In the CNN part, all the three components (residual module, ROI attention, and skip attention) were approximately equally important to the framework performance; removing one of them resulted in a 3-5% decline in performance. The proposed CNN improved ~4% overall performance and ~350% computational efficiency, compared to the U-Net model. The deep convolutional neural network, in conjunction with image processing for feature enhancement, is capable of feature extraction from CT images for pulmonary perfusion synthesis. In the proposed framework, image processing, especially CT contrast enhancement, plays a crucial role in the perfusion synthesis. This CTPM framework provides insights for relevant research studies in the future and enables other researchers to leverage for the development of optimized CNN models for functional lung avoidance radiation therapy.

Keywords: CT based image analysis; deep learning; functional lung avoidance radiation therapy; lung function imaging; perfusion imaging; perfusion synthesis.

Figure 1

The workflow of this framework. The effects of the components in the dashed outlined box were evaluated using ablation experiments.

Figure 2

The architecture of the proposed CNN model. The 3D blocks indicate the feature map. StrideConv was short for stride convolution. ResConv is short for residual convolution. ConvT is short for convolution transpose. Sigmoid is short for the sigmoid layer. Skip Att is short for the skip attention module. ROI Att is short for ROI attention.

Figure 3

Illustration of the effects of different image processing steps in coronal view. Two representative cases with the sharp low functional and gradient low functional were visualized for comparison. The SPECT perfusion was normalized using the approach described in the method section. The color means the perfusion level. SPECT-disc. was the discretized label. CTPM was generated using the proposed setting. The following CTPMs were generated with five ablation scenarios: disc.—remove label discretization; contr.—remove CT contrast enhancement; med. —remove the median filter in CT low functional enhancement; uni.—remove the uniform filter in CT low functional enhancement.

Figure 4

Illustration of the effect of varying CNN components and configurations of the representing case in coronal view. CTPM was generated using the proposed setting. The following CTPMs were generated in different CNN components and configurations: ROI.—remove ROI attention module, Skip.—remove skip attention module; Res.—remove residual module; ReL. – use ReLU instead of PReLU; LRe. – use LReLU instead of PReLU; wid.—with CNN width of 3, 4, or 5 times; dro—with a dropout rate of 0, 0.2, or 0.3; ker 3×—use kernel size of 3 × 3 × 3 install of 5 × 5 × 5.

Figure 5

Overall performance analysis of the CTPM framework for the testing group. (A) DSC of the high functional lung. (B) DSC of the low functional lung. (C) Correlation coefficient. (D) Structural similarity. CTPM_{-w/o pro} indicates prediction without image processing. CTPM_-U-Net indicates prediction using U-Net.

Figure 6

Comparison of the SPECT perfusion and CTPM perfusion of representative cases in the testing group. The arrows point to the primary low functional regions.

Figure 7

Illustration of the CT contrast enhancement of a representative case in coronal view. The CT image was visualized in the range of −1000 to 0. The SPECT image was visualized in the range of 0 to 800. CT_{-w/o contr} is the CT image after contrast enhancement.

Figure 8

Effects of the SPECT discretization. CTPM_{-w/o disc} indicates prediction without SPECT discretization procedure.