Detectability of Hypoattenuating Liver Lesions with Deep Learning CT Reconstruction: A Phantom and Patient Study

Abstract

Background CT deep learning image reconstruction (DLIR) improves image quality by reducing noise compared with adaptive statistical iterative reconstruction-V (ASIR-V). However, objective assessment of low-contrast lesion detectability is lacking. Purpose To investigate low-contrast detectability of hypoattenuating liver lesions on CT scans reconstructed with DLIR compared with CT scans reconstructed with ASIR-V in a patient and a phantom study. Materials and Methods This single-center retrospective study included patients undergoing portal venous phase abdominal CT between February and May 2021 and a low-contrast-resolution phantom scanned with the same protocol. Four reconstructions (ASIR-V at 40% strength [ASIR-V 40] and DLIR at three strengths) were generated. Five radiologists qualitatively assessed the images using the five-point Likert scale for image quality, lesion diagnostic confidence, conspicuity, and small lesion (≤1 cm) visibility. Up to two key lesions per patient, confirmed at histopathologic testing or at prior or follow-up imaging studies, were included. Lesion-to-background contrast-to-noise ratio was calculated. Interreader variability was analyzed. Intergroup qualitative and quantitative metrics were compared between DLIR and ASIR-V 40 using proportional odds logistic regression models. Results Eighty-six liver lesions (mean size, 15 mm ± 9.5 [SD]) in 50 patients (median age, 62 years [IQR, 57-73 years]; 27 [54%] female patients) were included. Differences were not detected for various qualitative low-contrast detectability metrics between ASIR-V 40 and DLIR (P > .05). Quantitatively, medium-strength DLIR and high-strength DLIR yielded higher lesion-to-background contrast-to-noise ratios than ASIR-V 40 (medium-strength DLIR vs ASIR-V 40: odds ratio [OR], 1.96 [95% CI: 1.65, 2.33]; high-strength DLIR vs ASIR-V 40: OR, 5.36 [95% CI: 3.68, 7.82]; P < .001). Low-contrast lesion attenuation was reduced by 2.8-3.6 HU with DLIR. Interreader agreement was moderate to very good for the qualitative metrics. Subgroup analysis based on lesion size of larger than 1 cm and 1 cm or smaller yielded similar results (P > .05). Qualitatively, phantom study results were similar to those in patients (P > .05). Conclusion The detectability of low-contrast liver lesions was similar on CT scans reconstructed with low-, medium-, and high-strength DLIR and ASIR-V 40 in both patient and phantom studies. Lesion-to-background contrast-to-noise ratios were higher for DLIR medium- and high-strength reconstructions compared with ASIR-V 40. © RSNA, 2024 Supplemental material is available for this article.

Graphical abstract

Figure 1:

Study flowchart shows details of patient recruitment, including the inclusion and exclusion criteria, the various CT reconstruction methods used, and the qualitative and quantitative analyses performed. ASIR-V 40% = adaptive statistical iterative reconstruction-V at 40% strength, DLIR-H = high-strength deep learning image reconstruction, DLIR-L = low-strength deep learning image reconstruction, DLIR-M = medium-strength deep learning image reconstruction, HIPAA = Health Insurance Portability and Accountability Act, SE = single energy.

Figure 2:

Phantom study flowchart shows a phantom (Catphan 604 CTP730; The Phantom Laboratory) containing 27 low-contrast hypoattenuating objects divided into three groups of nominal object contrast levels (group 1, 0.3%; group 2, 0.5%; group 3, 1%), each containing nine different-sized rods (size range, 2–15 mm). The labeling of the phantom was based on the Catphan manual. The phantom was scanned with the same abdominal CT protocol, and four reconstructions (adaptive statistical iterative reconstruction-V at 40% strength, low-strength deep learning image reconstruction [DLIR], medium-strength DLIR, and high-strength DLIR) were obtained in the scanner console. Five readers qualitatively analyzed the phantom data for overall image quality, object counting in each group, diagnostic confidence, and conspicuity of 15- and 8-mm objects in each group, and subcentimeter visibility of an 8-mm object in each group.

Figure 3:

Qualitative analysis of deidentified patient data. (A) Predefined screen layout shows images presented to the reader: a snapshot of key lesion 1 (dashed white circle) in segment VI (left side), a snapshot of key lesion 2 (dashed yellow circle) in segment VII (middle), and a series of axial CT images for the reader to scroll, pan, measure, or adjust the window levels for optimal assessment (right side). (B) Axial contrast-enhanced CT representative images of four reconstructions (adaptive statistical iterative reconstruction–V at 40% strength [ASIR-V 40], low-strength deep learning image reconstruction [DLIR-L], medium-strength DLIR [DLIR-M], and high-strength DLIR [DLIR-H]) in the same patient show the key lesion in segment VIII of the liver (arrows). The window width is 400 HU and the level is 40 HU.

Figure 4:

Two clinical examples used for qualitative results in the patient study with abdominal portal venous phase acquisition at single-energy CT. Axial images (top row: a key lesion in segment V [arrows]; bottom row: a key lesion in segment V [arrows]) show similar low-contrast detectability between adaptive statistical iterative reconstruction-V at 40% strength (ASIR-V 40) and three strengths of deep learning image reconstruction (DLIR) with improved image noise. DLIR-H = high-strength DLIR, DLIR-L = low-strength DLIR, M-DLIR = medium-strength DLIR.

Figure 5:

Plots show the qualitative scores for each reader in the phantom study. The x-axis represents the reconstruction method and the y-axis represents the frequency of the ranking, with the median score of five readers depicted by a blue dot in each graph. (A) Plot shows the overall image quality score for each reconstruction method depicted. (B) Plot shows the number of objects detected in each group. (C) Plot shows the diagnostic confidence of 8- and 15-mm objects. (D) Plot shows the conspicuity of 8- and 15-mm objects. (E) Plot shows the subcentimeter visibility of an 8-mm object for each group of contrast level. ASIR-V 40 = adaptive statistical iterative reconstruction-V at 40% strength, DLIR-H = high-strength deep learning image reconstruction, DLIR-L = low-strength deep learning image reconstruction, DLIR-M = medium-strength deep learning image reconstruction.

Figure 6:

Plots show qualitative comparisons of different reconstructions for standard- and low-dose CT in the phantom study. The x-axis represents the reconstruction method along with dose, and the y-axis represents the frequency of the ranking, with the median score of five readers depicted by a blue dot on each graph. (A) Plot shows the overall image quality score for adaptive statistical iterative reconstruction-V at 40% strength (A-40) standard (std) dose versus low-strength deep learning image reconstruction (D-L), medium-strength deep learning image reconstruction (D-M), and high-strength deep learning image reconstruction (D-H) low-dose reconstructions at 75%, 50%, and 25% of standard dose level. (B) Plot shows the number of objects detected in each group according to contrast level. (C) Plot shows the diagnostic confidence of 8-mm and 15-mm objects. (D) Plot shows the conspicuity of 8- and 15-mm objects. (E) Graph shows the subcentimeter visibility of an 8-mm object for each group according to contrast level.