Dataset Information

Devising a deep neural network based mammography phantom image filtering algorithm using images obtained under mAs and kVp control.

ABSTRACT: We study whether deep neural network based algorithm can filter out mammography phantom images that will pass or fail. With 543 phantom images generated from a mammography unit, we created VGG16 based phantom shape scoring models (multi-and binary-class classifiers). Using these models we designed filtering algorithms that can filter failed or passed phantom images. 61 phantom images obtained from two different medical institutions were used for external validation. The performances of the scoring models show an F1-score of 0.69 (95% confidence interval (CI) 0.65, 0.72) for multi-class classifiers and an F1-score of 0.93 (95% CI 0.92, 0.95) and area under the receiver operating characteristic curve of 0.97 (95% CI 0.96, 0.98) for binary-class classifiers. A total of 42 of the 61 phantom images (69%) were filtered by the filtering algorithms without further need for assessment from a human observer. This study demonstrated the potential to reduce the human workload from mammographic phantom interpretation using the deep neural network based algorithm.

SUBMITTER: Park SS

PROVIDER: S-EPMC9981722 | biostudies-literature | 2023 Mar

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Devising a deep neural network based mammography phantom image filtering algorithm using images obtained under mAs and kVp control.

Park Sung Soo SS Ku Young Mi YM Seo Kyung Jin KJ Whang In Yong IY Hwang Yun Sup YS Kim Min Ji MJ Jung Na Young NY

Scientific reports 20230302 1

We study whether deep neural network based algorithm can filter out mammography phantom images that will pass or fail. With 543 phantom images generated from a mammography unit, we created VGG16 based phantom shape scoring models (multi-and binary-class classifiers). Using these models we designed filtering algorithms that can filter failed or passed phantom images. 61 phantom images obtained from two different medical institutions were used for external validation. The performances of the scori ...[more]

PMID: 36864167

Similar Datasets

Project description:ObjectiveTo investigate the accuracy of the Agatston score obtained with the ultra-high-pitch (UHP) acquisition mode using tin-filter spectral shaping (Sn150 kVp) and a kVp-independent reconstruction algorithm to reduce the radiation dose.Materials and methodsThis prospective study included 114 patients (mean ± standard deviation, 60.3 ± 9.8 years; 74 male) who underwent a standard 120 kVp scan and an additional UHP Sn150 kVp scan for coronary artery calcification scoring (CACS). These two datasets were reconstructed using a standard reconstruction algorithm (120 kVp + Qr36d, protocol A; Sn150 kVp + Qr36d, protocol B). In addition, the Sn150 kVp dataset was reconstructed using a kVp-independent reconstruction algorithm (Sn150 kVp + Sa36d, protocol C). The Agatston scores for protocols A and B, as well as protocols A and C, were compared. The agreement between the scores was assessed using the intraclass correlation coefficient (ICC) and the Bland-Altman plot. The radiation doses for the 120 kVp and UHP Sn150 kVp acquisition modes were also compared.ResultsNo significant difference was observed in the Agatston score for protocols A (median, 63.05; interquartile range [IQR], 0-232.28) and C (median, 60.25; IQR, 0-195.20) (p = 0.060). The mean difference in the Agatston score for protocols A and C was relatively small (-7.82) and with the limits of agreement from -65.20 to 49.56 (ICC = 0.997). The Agatston score for protocol B (median, 34.85; IQR, 0-120.73) was significantly underestimated compared with that for protocol A (p < 0.001). The UHP Sn150 kVp mode facilitated an effective radiation dose reduction by approximately 30% (0.58 vs. 0.82 mSv, p < 0.001) from that associated with the standard 120 kVp mode.ConclusionThe Agatston scores for CACS with the UHP Sn150 kVp mode with a kVp-independent reconstruction algorithm and the standard 120 kVp demonstrated excellent agreement with a small mean difference and narrow agreement limits. The UHP Sn150 kVp mode allowed a significant reduction in the radiation dose.

Project description:BACKGROUND:Two novel methods of image reconstruction, xSPECT Quant (xQ) and xSPECT Bone (xB), that use an ordered subset conjugate gradient minimizer (OSCGM) for SPECT/CT reconstruction have been proposed. The present study compares the performance characteristics of xQ, xB, and conventional Flash3D (F3D) reconstruction using images derived from phantoms and patients. METHODS:A custom-designed body phantom for bone SPECT was scanned using a Symbia Intevo (Siemens Healthineers), and reconstructed xSPECT images were evaluated. The phantom experiments proceeded twice with different activity concentrations and sphere sizes. A phantom with 28-mm spheres containing a 99mTc-background and tumor-to-normal bone ratios (TBR) of 1, 2, 4, and 10 were generated, and convergence property against various TBR was evaluated across 96 iterations. A phantom with four spheres (13-, 17-, 22-, and 28-mm diameters), containing a 99mTc-background at TBR4, was also generated. The full width at half maximum of an imaged spinous process (10?mm), coefficients of variance (CV), contrast-to-noise ratio (CNR), and recovery coefficients (RC) were evaluated after reconstructing images of a spine using Flash 3D (F3D), xQ, and xB. We retrospectively analyzed images from 20 patients with suspected bone metastases (male, n = 13) which were acquired using [99mTc]Tc-(H)MDP SPECT/CT, then CV and standardized uptake values (SUV) at the 4th vertebral body (L4) were compared after xQ and xB reconstruction in a clinical setup. RESULTS:Mean activity concentrations with various TBR converged according to increasing numbers of iterations. The spatial resolution of xB was considerably superior to xQ and F3D, and it approached almost the actual size regardless of the iteration numbers during reconstruction. The CV and RC were better for xQ and xB than for F3D. The CNR peaked at 24 iterations for xQ and 48 iterations for F3D and xB, respectively. The RC between xQ and xB significantly differed at lower numbers of iterations but were almost equivalent at higher numbers of iterations. The reconstructed xQ and xB images of the clinical patients showed a significant difference in the SUVmax and SUVpeak. CONCLUSIONS:The reconstructed xQ and xB images were more accurate than those reconstructed conventionally using F3D. The xB for bone SPECT imaging offered essentially unchanged spatial resolution even when the numbers of iterations did not converge. The xB reconstruction further enhanced SPECT image quality using CT data. Our findings provide important information for understanding the performance characteristics of the novel xQ and xB algorithms.

Project description:PurposeThe purpose of this study was to evaluate the image quality and radiation dose in computed tomography urography (CTU) images acquired with a low kilovoltage peak (kVp) in combination with an adaptive statistical iterative reconstruction (ASiR) algorithm.MethodsA total of 45 subjects (18 women, 27 men) who underwent CTU with kV assist software for automatic selection of the optimal kVp were included and divided into two groups (A and B) based on the kVp and image reconstruction algorithm: group A consisted of patients who underwent CTU with a 80 or 100 kVp and whose images were reconstructed with the 50% ASiR algorithm (n=32); group B consisted of patients who underwent CTU with a 120 kVp and whose images were reconstructed with the filtered back projection (FBP) algorithm (n=13). The images were separately reconstructed with volume rendering (VR) and maximum intensity projection (MIP). Finally, the image quality was evaluated using an image score, CT attenuation, image noise, the contrast-to-noise ratio (CNR) of the renal pelvis-to-abdominal visceral fat and the signal-to-noise ratio (SNR) of the renal pelvis. The radiation dose was assessed using volume CT dose index (CTDIvol), dose-length product (DLP) and effective dose (ED).ResultsFor groups A and B, the subjective image scores for the VR reconstruction images were 3.9±0.4 and 3.8±0.4, respectively, while those for the MIP reconstruction images were 3.8±0.4 and 3.6±0.6, respectively. No significant difference was found (p>0.05) between the two groups' image scores for either the VR or MIP reconstruction images. Additionally, the inter-reviewer image scores did not significantly differ (p>0.05). The mean attenuation of the bilateral renal pelvis in group A was significantly higher than that in group B (271.4±57.6 vs. 221.8±35.3 HU, p<0.05), whereas the image noise in group A was significantly lower than that in group B (7.9±2.1 vs. 10.5±2.3 HU, p<0.05). The CNR and SNR in group A were both significantly higher than those in group B (53.61±24.74 vs. 32.30±6.52 for CNR; 38.13±19.86 vs. 21.76±4.85 for SNR; all p<0.05). The CTDIvol, DLP and ED in group A were significantly lower than those in group B (9.26±2.77 vs. 16.19±5.60 mGy for CTDIvol; 368.86±119.38 vs. 674.38±239.37 mGy×cm-1 for DLP; 5.53±1.79 vs. 10.12±3.59 mSv for ED; all p<0.05).ConclusionsThe low kVp CTU images with 50% ASiR reconstruction exhibit sufficient image quality and facilitate up to a 44% radiation dose reduction.

Dataset Information

Devising a deep neural network based mammography phantom image filtering algorithm using images obtained under mAs and kVp control.

Publications

Devising a deep neural network based mammography phantom image filtering algorithm using images obtained under mAs and kVp control.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets