Deep learning reconstruction for detection of liver lesions at standard-dose and reduced-dose abdominal CT

OBJECTIVES

Deep learning reconstruction (DLR) has shown promising image denoising ability, but its radiation dose reduction potential remains unknown. The objective of this study was to investigate the diagnostic performance of DLR compared to iterative reconstruction (IR) in the detection of liver lesions at standard-dose and reduced-dose CT.

MATERIALS AND METHODS

Participants with known liver metastases from gastrointestinal and pancreatic adenocarcinoma were prospectively included from routine follow-up (October 2020 to March 2022). Participants received standard-dose CT and two additional reduced-dose scans during the same contrast administration, each reconstructed with IR and high-strength DLR. Two radiologists evaluated images for the presence of liver lesions, and a third established a reference standard. Diagnostic performance was compared using McNemar's test and mixed effects logistic regression.

RESULTS

Forty-four participants (mean age 66 years ± 11 [standard deviation], 28 men) were evaluated with 348 included liver lesions ≤ 20 mm (297 metastases, 51 benign; mean size 9.1 ± 4.3 mm). Mean volume CT dose index was 14.2, 7.8 mGy, and 5.1 mGy. Between algorithms, no significant difference in lesion detection was observed within dose levels. Detection of 233 lesions ≤ 10 mm was deteriorated with lower dose levels despite DLR denoising, with 185 detected at standard-dose IR (79.4%; 95% CI: 73.5-84.3) vs 128 at medium-dose DLR (54.9%; 95% CI: 48.3-61.4; p < 0.001) and 105 at low-dose DLR (45.1%; 95% CI: 38.6-51.7; p < 0.001).

CONCLUSION

Diagnostic performance for liver lesion detection was comparable between algorithms. When the detection of smaller lesions is important, DLR did not facilitate substantial dose reduction.

KEY POINTS

Question Methods to reduce CT radiation dose are desirable in clinical practice, and DLR has shown promising image denoising capabilities. Findings Liver lesion detection was comparable for DLR and IR across dose levels, but detection of smaller lesions deteriorated with lower dose levels. Clinical relevance Although potent in image noise reduction, the diagnostic performance of DLR is comparable to IR at standard-dose and reduced-dose CT. Care must be taken in pursuit of dose reduction when the detection and characterization of smaller liver lesions are of clinical importance.

Size and Contrast Thresholds for Liver Lesion Detection in Regular and Low-dose CT Examinations: A Reader Study of 2300 Synthetic Lesions Across 100 Patients

RATIONALE AND OBJECTIVES

To determine the size and contrast required for liver lesion detection in regular and low-dose computed tomography (CT) examinations.

MATERIALS AND METHODS

100 abdominal CT datasets were retrospectively collected, with 50 originating from vendor A and 50 from vendor B. Half the datasets from each scanner were regular-dose oncologic examinations, the other half were acquired using a low-dose kidney stone protocol. Cylindrical liver lesions with 23 different combinations of diameter and contrast to the surrounding liver were digitally inserted. Seven radiologists assessed lesion detectability in a four-alternative forced choice reading experiment, and image noise was measured within the liver.

RESULTS

Lesion detection thresholds at regular dose were at -30, -35, and -70 Hounsfield unit (HU) lesion contrast (vendor A) and -25, -35, and -65 HU (vendor B) for lesions with 15-, 10-, and 5-mm diameter, respectively. At low dose, thresholds were -40 and -45 HU (vendor A) and -40 and -50 HU (vendor B) for 15- and 10-mm lesions, while 5-mm lesions did not reach the detection threshold. Noise levels were 21.5±2.3 HU at regular dose vs 22.2±2.0 HU at low dose for vendor A (P=.06) and 25.9±4.9 HU vs 30.9±3.1 HU for vendor B (P<.001).

CONCLUSION

In oncologic CT examinations, liver lesions with diameters of 15-, 10-, and 5-mm require contrasts of -30, -35, and -70 HU, respectively for reliable detection. In low-dose examinations, greater contrasts of -40 and -50 HU are required for lesions measuring 15- and 10-mm, while readers do not reliably detect 5-mm lesions.

Evaluation of Image Quality and Detectability of Deep Learning Image Reconstruction (DLIR) Algorithm in Single- and Dual-energy CT

This study is aimed to evaluate effects of deep learning image reconstruction (DLIR) on image quality in single-energy CT (SECT) and dual-energy CT (DECT), in reference to adaptive statistical iterative reconstruction-V (ASIR-V). The Gammex 464 phantom was scanned in SECT and DECT modes at three dose levels (5, 10, and 20 mGy). Raw data were reconstructed using six algorithms: filtered back-projection (FBP), ASIR-V at 40% (AV-40) and 100% (AV-100) strength, and DLIR at low (DLIR-L), medium (DLIR-M), and high strength (DLIR-H), to generate SECT 120kVp images and DECT 120kVp-like images. Objective image quality metrics were computed, including noise power spectrum (NPS), task transfer function (TTF), and detectability index (d'). Subjective image quality evaluation, including image noise, texture, sharpness, overall quality, and low- and high-contrast detectability, was performed by six readers. DLIR-H reduced overall noise magnitudes from FBP by 55.2% in a more balanced way of low and high frequency ranges comparing to AV-40, and improved the TTF values at 50% for acrylic inserts by average percentages of 18.32%. Comparing to SECT 20 mGy AV-40 images, the DECT 10 mGy DLIR-H images showed 20.90% and 7.75% improvement in d' for the small-object high-contrast and large-object low-contrast tasks, respectively. Subjective evaluation showed higher image quality and better detectability. At 50% of the radiation dose level, DECT with DLIR-H yields a gain in objective detectability index compared to full-dose AV-40 SECT images used in daily practice.

Assessment of low-dose paranasal sinus CT imaging using a new deep learning image reconstruction technique in children compared to adaptive statistical iterative reconstruction V (ASiR-V)

Purpose

To compare the effects of deep learning image reconstruction (DLIR) and adaptive statistical iterative reconstruction V (ASiR-V) on image quality in low-dose computed tomography (CT) of paranasal sinuses in children.

Methods

Low-dose CT scans of the paranasal sinuses in 25 pediatric patients were retrospectively evaluated. The raw data were reconstructed with three levels of DLIR (high, H; medium, M; and low, L), filtered back projection (FBP), and ASiR-V (30% and 50%). Image noise was measured in both soft tissue and bone windows, and the signal-to-noise ratios (SNRs) and contrast-to-noise ratios (CNRs) of the images were calculated. Subjective image quality at the ethmoid sinus and nasal cavity levels of the six groups of reconstructed images was assessed by two doctors using a five-point Likert scale in a double-blind manner.

Results

The patients’ mean dose-length product and effective dose were 36.65 ± 2.44 mGy·cm and 0.17 ± 0.03 mSv, respectively. (1) Objective evaluation: 1. Soft tissue window: The difference among groups in each parameter was significant (P < 0.05). Pairwise comparisons showed that the H group’ s parameters were significantly better (P < 0.05) than those of the 50% post-ASiR-V group. 2. Bone window: No significant between-group differences were found in the noise of the petrous portion of the temporal bone or its SNR or in the noise of the pterygoid processes of the sphenoids or their SNRs (P > 0.05). Significant differences were observed in the background noise and CNR (P < 0.05). As the DLIR intensity increased, image noise decreased and the CNR improved. The H group exhibited the best image quality. (2) Subjective evaluation: Scores for images of the ethmoid sinuses were not significantly different among groups (P > 0.05). Scores for images of the nasal cavity were significantly different among groups (P < 0.05) and were ranked in descending order as follows: H, M, L, 50% post-ASiR-V, 30% post-ASiR-V, and FBP.

Conclusion

DLIR was superior to FBP and post-ASiR-V in low-dose CT scans of pediatric paranasal sinuses. At high intensity (H), DLIR provided the best reconstruction effects.

The influence of a deep learning image reconstruction algorithm on the image quality and auto-analysis of pulmonary nodules at ultra-low dose chest CT: a phantom study

BACKGROUND

To investigate the effect of a new deep learning image reconstruction (DLIR) algorithm on the detection, characterization and image quality of pulmonary nodules (PNs) in ultra-low dose chest computed tomography (CT) in comparison with the adaptive statistical iterative reconstruction (ASIR-V) algorithm.

METHODS

Nine artificial pulmonary nodules [six ground glass nodules (GGNs) and three solid nodules (SNs); density: -800 HU, -630 HU, 100 HU; diameter: 12 mm, 10 mm, 8 mm] were randomly placed in a thorax anthropomorphic phantom (Lungman, Kyoto Kagaku Inc.) and scanned on a 256-row CT (Revolution CT, GE Healthcare). Eight scans were performed at 70 kVp with different tube currents (20, 30, 50, 70, 90, 100, 120, 150 mA). Raw data were reconstructed using the filtered back projection (FBP), ASIR-V (30%, 50%, 80%) and DLIR (Low, Medium, High; TrueFidelity™) at 0.625 mm thickness. The effective radiation dose was recorded. All images were automatically analyzed using a commercially available artificial intelligence software (Intelligent 4D Imaging System for Chest CT 5.5, YITU Healthcare) and CT value, standard deviation (SD), long and short diameters of each nodule and SD of air (background) were measured. The detection rate, deformation degree (long diameter/short diameter), signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of pulmonary nodules were calculated.

RESULTS

Nodule CT values were the same in all mA settings for all three types of reconstruction algorithms (all P>0.05). DLIR groups had significantly lower SD and higher SNR and CNR values, with better overall image quality than ASIR-V and FBP groups at each mA, ranging from 65-85% reduction in SD, 67-83% increase in SNR with DLIR-H over 50%ASIR-V and 75-91% reduction in SD and 77-89% increase in SNR with DLIR-H over FBP (all P<0.05). At ultra-low dose conditions (30 mA), the DLIR-H images had the highest detection rate of PNs (100%). In addition, the DLIR-M had a minimal negative effect on the characterization of PNs.

CONCLUSIONS

DLIR algorithm can be a potential reconstruction technique to optimize image quality and improve detection rate of PNs in ultra-low dose lung screening.

Reduced-Dose Deep Learning Reconstruction for Abdominal CT of Liver Metastases

Background Assessment of liver lesions is constrained as CT radiation doses are lowered; evidence suggests deep learning reconstructions mitigate such effects. Purpose To evaluate liver metastases and image quality between reduced-dose deep learning image reconstruction (DLIR) and standard-dose filtered back projection (FBP) contrast-enhanced abdominal CT. Materials and Methods In this prospective Health Insurance Portability and Accountability Act-compliant study (September 2019 through April 2021), participants with biopsy-proven colorectal cancer and liver metastases at baseline CT underwent standard-dose and reduced-dose portal venous abdominal CT in the same breath hold. Three radiologists detected and characterized lesions at standard-dose FBP and reduced-dose DLIR, reported confidence, and scored image quality. Contrast-to-noise ratios for liver metastases were recorded. Summary statistics were reported, and a generalized linear mixed model was used. Results Fifty-one participants (mean age ± standard deviation, 57 years ± 13; 31 men) were evaluated. The mean volume CT dose index was 65.1% lower with reduced-dose CT (12.2 mGy) than with standard-dose CT (34.9 mGy). A total of 161 lesions (127 metastases, 34 benign lesions) with a mean size of 0.7 cm ± 0.3 were identified. Subjective image quality of reduced-dose DLIR was superior to that of standard-dose FBP (P < .001). The mean contrast-to-noise ratio for liver metastases of reduced-dose DLIR (3.9 ± 1.7) was higher than that of standard-dose FBP (3.5 ± 1.4) (P < .001). Differences in detection were identified only for lesions 0.5 cm or smaller: 63 of 65 lesions detected with standard-dose FBP (96.9%; 95% CI: 89.3, 99.6) and 47 lesions with reduced-dose DLIR (72.3%; 95% CI: 59.8, 82.7). Lesion accuracy with standard-dose FBP and reduced-dose DLIR was 80.1% (95% CI: 73.1, 86.0; 129 of 161 lesions) and 67.1% (95% CI: 59.3, 74.3; 108 of 161 lesions), respectively (P = .01). Lower lesion confidence was reported with a reduced dose (P < .001). Conclusion Deep learning image reconstruction (DLIR) improved CT image quality at 65% radiation dose reduction while preserving detection of liver lesions larger than 0.5 cm. Reduced-dose DLIR demonstrated overall inferior characterization of liver lesions and reader confidence. Clinical trial registration no. NCT03151564 © RSNA, 2022 Online supplemental material is available for this article.

Deep-learning-based image reconstruction in dynamic contrast-enhanced abdominal CT: image quality and lesion detection among reconstruction strength levels

AIM

To evaluate the use of deep-learning-based image reconstruction (DLIR) algorithms in dynamic contrast-enhanced computed tomography (CT) of the abdomen, and to compare the image quality and lesion conspicuity among the reconstruction strength levels.

MATERIALS AND METHODS

This prospective study included 59 patients with 373 hepatic lesions who underwent dynamic contrast-enhanced CT of the abdomen. All images were reconstructed using four reconstruction algorithms, including 40% adaptive statistical iterative reconstruction-Veo (ASiR-V) and DLIR at low, medium, and high-strength levels (DLIR-L, DLIR-M, and DLIR-H, respectively). The signal-to-noise ratio (SNR) of the abdominal aorta, portal vein, liver, pancreas, and spleen and the lesion-to-liver contrast-to-noise ratio (CNR) were calculated and compared among the four reconstruction algorithms. The diagnostic acceptability was qualitatively assessed and compared among the four reconstruction algorithms and the conspicuity of hepatic lesions was compared between <5 and ≥5 mm lesions.

RESULTS

The SNR of each anatomical structure (p<0.0001) and CNR (p<0.0001) were significantly higher in DLIR-H than the other reconstruction algorithms. Diagnostic acceptability was significantly better in DLIR-M than the other reconstruction algorithms (p<0.0001). The conspicuity of hepatic lesions was highest when using 40% ASiR-V and tended to lessen as the reconstruction strength level was getting higher in DLIR, especially in <5 mm lesions; however, all hepatic lesions could be detected.

CONCLUSIONS

DLIR improved the SNR, CNR, and image quality compared with 40% ASiR-V, while making it possible to decrease lesion conspicuity using higher reconstruction strength.

Impact of novel deep learning image reconstruction algorithm on diagnosis of contrast-enhanced liver computed tomography imaging: Comparing to adaptive statistical iterative reconstruction algorithm

OBJECTIVE

To assess clinical application of applying deep learning image reconstruction (DLIR) algorithm to contrast-enhanced portal venous phase liver computed tomography (CT) for improving image quality and lesions detection rate compared with using adaptive statistical iterative reconstruction (ASIR-V) algorithm under routine dose.

METHODS

The raw data from 42 consecutive patients who underwent contrast-enhanced portal venous phase liver CT were reconstructed using three strength levels of DLIRs (low [DL-L]; medium [DL-M]; high [DL-H]) and two levels of ASIR-V (30%[AV-30]; 70%[AV-70]). Objective image parameters, including noise, signal-to-noise (SNR), and the contrast-to-noise ratio (CNR) relative to muscle, as well as subjective parameters, including noise, artifact, hepatic vein-clarity, index lesion-clarity, and overall scores were compared pairwise. For the lesions detection rate, the five reconstructions in patients who underwent subsequent contrast-enhanced magnetic resonance imaging (MRI) examinations were compared.

RESULTS

For objective parameters, DL-H exhibited superior image quality of lower noise and higher SNR than AV-30 and AV-70 (all P < 0.05). CNR was not statistically different between AV-70, DL-M, and DL-H (all P > 0.05). In both objective and subjective parameters, only image noise was statistically reduced as the strength of DLIR increased compared with ASIR-V (all P < 0.05). Regarding the lesions detection rate, a total of 45 lesions were detected by MRI examination and all five reconstructions exhibited similar lesion-detection rate (25/45, 55.6%).

CONCLUSION

Compared with AV-30 and AV 70, DLIR leads to better image quality with equal lesion detection rate for liver CT imaging under routine dose.

Image Quality Assessment of Abdominal CT by Use of New Deep Learning Image Reconstruction: Initial Experience

OBJECTIVE. The purpose of this study was to perform quantitative and qualitative evaluation of a deep learning image reconstruction (DLIR) algorithm in contrast-enhanced oncologic CT of the abdomen. MATERIALS AND METHODS. Retrospective review (April-May 2019) of the cases of adults undergoing oncologic staging with portal venous phase abdominal CT was conducted for evaluation of standard 30% adaptive statistical iterative reconstruction V (30% ASIR-V) reconstruction compared with DLIR at low, medium, and high strengths. Attenuation and noise measurements were performed. Two radiologists, blinded to examination details, scored six categories while comparing reconstructions for overall image quality, lesion diagnostic confidence, artifacts, image noise and texture, lesion conspicuity, and resolution. RESULTS. DLIR had a better contrast-to-noise ratio than 30% ASIR-V did; high-strength DLIR performed the best. High-strength DLIR was associated with 47% reduction in noise, resulting in a 92-94% increase in contrast-to-noise ratio compared with that of 30% ASIR-V. For overall image quality and image noise and texture, DLIR scored significantly higher than 30% ASIR-V with significantly higher scores as DLIR strength increased. A total of 193 lesions were identified. The lesion diagnostic confidence, conspicuity, and artifact scores were significantly higher for all DLIR levels than for 30% ASIR-V. There was no significant difference in perceived resolution between the reconstruction methods. CONCLUSION. Compared with 30% ASIR-V, DLIR improved CT evaluation of the abdomen in the portal venous phase. DLIR strength should be chosen to balance the degree of desired denoising for a clinical task relative to mild blurring, which increases with progressively higher DLIR strengths.