Evaluation of a Generative Adversarial Network to Improve Image Quality and Reduce Radiation-Dose during Digital Breast Tomosynthesis

Abstract

In this study, we evaluated the improvement of image quality in digital breast tomosynthesis under low-radiation dose conditions of pre-reconstruction processing using conditional generative adversarial networks [cGAN (pix2pix)]. Pix2pix pre-reconstruction processing with filtered back projection (FBP) was compared with and without multiscale bilateral filtering (MSBF) during pre-reconstruction processing. Noise reduction and preserve contrast rates were compared using full width at half-maximum (FWHM), contrast-to-noise ratio (CNR), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM) in the in-focus plane using a BR3D phantom at various radiation doses [reference-dose (automatic exposure control reference dose: AECrd), 50% and 75% reduction of AECrd] and phantom thicknesses (40 mm, 50 mm, and 60 mm). The overall performance of pix2pix pre-reconstruction processing was effective in terms of FWHM, PSNR, and SSIM. At ~50% radiation-dose reduction, FWHM yielded good results independently of the microcalcification size used in the BR3D phantom, and good noise reduction and preserved contrast. PSNR results showed that pix2pix pre-reconstruction processing represented the minimum in the error with reference FBP images at an approximately 50% reduction in radiation-dose. SSIM analysis indicated that pix2pix pre-reconstruction processing yielded superior similarity when compared with and without MSBF pre-reconstruction processing at ~50% radiation-dose reduction, with features most similar to the reference FBP images. Thus, pix2pix pre-reconstruction processing is promising for reducing noise with preserve contrast and radiation-dose reduction in clinical practice.

Keywords: digital breast tomosynthesis; generative adversarial networks; improve image quality; radiation-dose reduction.

Figure 1

Optimization results for parameter (epochs) determination for pix2pix pre-reconstruction processing at different radiation-dose levels: (a) 0.66 mGy (approximately 50% of automatic exposure control reference dose [AECrd]), (b) 0.31 mGy (approximately 50% reduction of AECrd), (c) 0.66 mGy (approximately 75% reduction of AECrd), and (d) 0.31 mGy (approximately 75% reduction of AECrd). BR3D phantom thickness: 40 mm.

Figure 2

Comparisons between pix2pix pre-reconstruction processing and conventional multiscale bilateral filtering (MSBF) pre-reconstruction processing with and without noise reduction (microcalcifications (MCs) [showing window: 0.41–0.70]; masses [3.9 mm: 0.41–0.70, 4.7 mm: 0.35–0.64]) in the in-focus plane (BR3D phantom thickness: 40 mm). The display referring to the image contrast of the BR3D phantom was changed for visual comparison of the signal and background gray levels. h: approximately 50% reduction of automatic exposure control reference dose (AECrd), q: approximately 75% reduction of AECrd.

Figure 3

Comparisons between pix2pix pre-reconstruction processing and conventional multiscale bilateral filtering (MSBF) pre-reconstruction processing with and without noise reduction (microcalcifications (MCs) [showing window: 0.39–0.68]; masses [3.9 mm: 0.39–0.68; 4.7 mm: 0.28–0.57]) in the in-focus plane (BR3D phantom thickness: 50 mm). The display referring to the image contrast of the BR3D phantom was changed for visual comparison of the signal and background gray levels. h: approximately 50% reduction of automatic exposure control reference dose (AECrd); q: approximately 75% reduction of AECrd.

Figure 4

Comparisons between pix2pix pre-reconstruction processing and the conventional multiscale bilateral filtering (MSBF) pre-reconstruction processing with and without noise reduction (microcalcifications (MCs) [showing window: 0.30–0.59]; masses [3.9 mm: 0.30–0.59; 4.7 mm: 0.23–0.52]) in the in-focus plane (BR3D phantom thickness; 60 mm). The display referring to the image contrast of the BR3D phantom was changed for visual comparison of the signal and background gray levels. h: approximately 50% reduction of automatic exposure control reference dose (AECrd); q: approximately 75% reduction of AECrd.

Figure 5

Comparisons of the full width at half-maximum (FWHM) determined for in-focus plane images obtained using pix2pix pre-reconstruction processing and reference, with and without multiscale bilateral filtering (MSBF) pre-reconstruction processing, for low dose with variant phantom thickness [microcalcifications (MCs) size: 0.19 mm]. (a) Vertical direction 40 mm; (b) vertical direction 50 mm; (c) vertical direction 60 mm; (d) horizontal direction 40 mm; (e) horizontal direction 50 mm; and (f) horizontal direction 60 mm. h: approximately 50% reduction of automatic exposure control reference dose (AECrd); q: approximately 75% reduction of AECrd.

Figure 6

Comparisons of the full width at half-maximum (FWHM) determined for in-focus plane images obtained using pix2pix pre-reconstruction processing and reference, with and without multiscale bilateral filtering (MSBF) pre-reconstruction processing, for low dose with variant phantom thickness [microcalcifications (MCs) size: 0.23 mm]. (a) Vertical direction 40 mm; (b) vertical direction 50 mm; (c) vertical direction 60 mm; (d) horizontal direction 40 mm; (e) horizontal direction 50 mm; and (f) horizontal direction 60 mm. h: approximately 50% reduction of automatic exposure control reference dose (AECrd); q: approximately 75% reduction of AECrd.

Figure 7

Comparisons of the full width at half-maximum (FWHM) determined for in-focus plane images obtained using pix2pix pre-reconstruction processing and reference, with and without multiscale bilateral filtering (MSBF) pre-reconstruction processing, for low dose with variant phantom thickness [microcalcifications (MCs) size: 0.29 mm]. (a) Vertical direction 40 mm; (b) vertical direction 50 mm; (c) vertical direction 60 mm; (d) horizontal direction 40 mm; (e) horizontal direction 50 mm; and (f) horizontal direction 60 mm. h: approximately 50% reduction of automatic exposure control reference dose (AECrd); q: approximately 75% reduction of AECrd.

Figure 8

Plots of the structural similarity (SSIM) vs. reference vs. without pre-reconstruction processing, reference vs. with multiscale bilateral filtering (MSBF) pre-reconstruction processing, reference vs. pix2pix pre-reconstruction processing from the in-focus plane for low dose with variant phantom thickness. (a) In-focus plane for evaluation of 0.19- and 0.23-mm microcalcifications (MCs); (b) in-focus plane for evaluation of 0.29-mm MCs and 3.9-mm masses; (c) in-focus plane for evaluation of 4.7-mm mass. h: approximately 50% reduction of automatic exposure control reference dose (AECrd); q: approximately 75% reduction of AECrd. (Tukey–Kramer test; p < 0.05 indicates a significant difference, *: significant).

Figure 9

Plots of the peak signal-to-noise ratio (PSNR) vs. reference vs. without pre-reconstruction processing, reference vs. with multiscale bilateral filtering (MSBF) pre-reconstruction processing, reference vs. pix2pix pre-reconstruction processing from the in-focus plane for low dose with variant phantom thickness. (a) In-focus plane for evaluation of 0.19- and 0.23-mm microcalcifications (MCs). (b) In-focus plane for evaluation of 0.29-mm MCs and 3.9-mm masses. (c) In-focus plane for evaluation of 4.7-mm mass. h: approximately 50% reduction of automatic exposure control reference dose (AECrd); q: approximately 75% reduction of AECrd. (Tukey–Kramer test; p < 0.05 indicates a significant difference, *: significant).

Figure 10

Plots of the contrast-to-noise ratio (CNR) vs. pre-reconstruction processing, with and without multiscale bilateral filtering (MSBF) pre-reconstruction processing from the in-focus plane. Comparisons of the CNR of the in-focus plane images obtained via the reference, low dose [approximately 50% and 25% of automatic exposure control reference dose (AECrd)] with and without pre-reconstruction processing with varying phantom thicknesses. (a) 4.7 mm mass; (b) 3.9 mm mass. The in-focus plane image shows the masses and background areas of the CNR. h: approximately 50% reduction of AECrd; q: approximately 75% reduction of AECrd. (Tukey–Kramer test; p < 0.05 indicates a significant difference, *: significant, NS: not significant).