Yu, Lifeng

Abstract This article provides an overview of deep-learning–based techniques in CT image reconstruction and processing (referred to as “DLR”), covering technical implementations, performance evaluation, radiation dose reduction, and future perspectives. DLR methods can be categorized into projection-space, projection-to-image-space, image-space, and various hybrid techniques, with applications such as noise reduction, artefact correction, and spatial resolution enhancement. Performance evaluations include phantom-based studies, patient image-based studies, and virtual imaging trials. These evaluation studies demonstrated that DLR can effectively reduce image noise while preserving an image texture like in traditional filtered-backprojection (FBP) images, although the extent of radiation dose reduction varies widely depending on the study and the specific diagnostic task. Challenges remain in low-contrast lesion detection and characterization, where dose reduction may still be less than 50% compared to traditional reconstruction methods. Additionally, the potential for DLR methods to generate false structures or “hallucinations,” especially at low radiation doses, emphasizes the need for effective monitoring and mitigation strategies from both technical and clinical perspectives. Quantitative, accurate, and efficient evaluation techniques, such as virtual image trial-based methods, can be explored to help optimize these algorithms for reducing radiation dose and enhancing diagnostic performance.

Objectives: To determine the feasibility of reduced-dose chest computed tomographic angiography (CTA) with convolutional neural network (CNN) denoising for detecting pulmonary arteriovenous malformations (pAVMs) in children with hereditary hemorrhagic telangiectasia (cwHHT). Methods: Fifteen cwHHT underwent a chest CTA (ie, a controlled “study” dose). Noise was inserted to simulate a quarter dose (QD) exam. Images were reconstructed using iterative reconstruction (IR) and our self-trained CNN denoising model. For each case, 3 sets of images were created: study dose (SD)+IR, QD+IR, and QD+CNN. Two thoracic radiologists independently scored each set to assess quality, spatial resolution, artifacts, and the presence of pAVMs using 4-level ordinal scales. Quantitative assessments of image quality were performed using contrast-to-noise ratios (CNRs) with comparisons made between the experimental conditions. Results: Thirteen of the 15 patients recruited with hereditary hemorrhagic telangiectasia (mean age: 9.3±4.5 y) were positive for pAVM by transthoracic contrast echocardiography. The sensitivities using QD+CNN were 0.85 and 1.00 for readers 1 and 2, respectively. This was compared with 0.69 and 0.84 using QD+IR versus 0.85 and 0.92 for SD+IR. Inter-reader agreement for pAVM detection utilizing QD+CNN was moderate and resulted in kappa=0.59 ( P =0.012). The subjective assessments for QD+CNN were comparable to the SD technique. Regression analysis of reader scores revealed improved quality in QD+CNN versus QD+IR ( P =0.001). Similarly, the QD+CNN condition demonstrated the highest CNRs. Conclusions: Reduced-dose chest CTA with CNN denoising provides a level of sensitivity comparable to standard dose CTA and high CNRs for the detection of pAVMs in cwHHT.

Abstract Objectives To develop a volume of interest (VOI) imaging technique in multi-detector-row helical CT to reduce radiation dose or improve image quality within the VOI. Methods A deep-learning method based on a residual U-Net architecture, named VOI-Net, was developed to correct truncation artifacts in VOI helical CT. Three patient cases, a chest CT of interstitial lung disease and 2 abdominopelvic CT of liver tumour, were used for evaluation through simulation. Results VOI-Net effectively corrected truncation artifacts (root mean square error [RMSE] of 5.97 ± 2.98 Hounsfield Units [HU] for chest, 3.12 ± 1.93 HU, and 3.71 ± 1.87 HU for liver). Radiation dose was reduced by 71% without sacrificing image quality within a 10-cm diameter VOI, compared to a full scan field of view (FOV) of 50 cm. With the same total energy deposited as in a full FOV scan, image quality within the VOI matched that at 350% higher radiation dose. A radiologist confirmed improved lesion conspicuity and visibility of small linear reticulations associated with ground-glass opacity and liver tumour. Conclusions Focusing radiation on the VOI and using VOI-Net in a helical scan, total radiation can be reduced or higher image quality equivalent to those at higher doses in standard full FOV scan can be achieved within the VOI. Advances in knowledge A targeted helical VOI imaging technique enabled by a deep-learning-based artifact correction method improves image quality within the VOI without increasing radiation dose.

Abstract Background Accurate low‐dose CT simulation is required to efficiently assess reconstruction and dose reduction techniques. Projection domain noise insertion requires proprietary information from manufacturers. Analytic image domain noise insertion methods are successful for linear reconstruction algorithms, however extending them to non‐linear algorithms remains challenging. Emerging, deep‐learning‐based image domain noise insertion methods have potential, but few approaches have explicitly incorporated physics information and a texture‐synthesis model to guide the generation of locally and globally correlated noise texture. Purpose We proposed a p hysics‐informed model‐based gener a tive neura l network for simulating scann e r‐ and algori t hm‐specific low‐dose C T e xams (PALETTE). It is expected to provide an alternative to projection domain noise insertion methods in the absence of manufacturers’ proprietary information and tools. Methods PALETTE integrated a physics‐based noise prior generation process, a Noise2Noisier sub‐network, and a noise texture synthesis sub‐network. The Noise2Noisier sub‐network provided a bias prior, which, combined with the noise prior, served as the inputs to noise texture synthesis sub‐network. Explicit regularizations in spatial and frequency domains were developed to account for noise spatial correlation and frequency characteristics. For proof‐of‐concept, PALETTE was trained and validated for a commercial iterative reconstruction algorithm (SAFIRE, Siemens Healthineers), using the paired routine and 25% dose images from CT phantoms (lateral size 30–40 cm; three training and four testing phantoms) and open‐access patient cases (10 training and 20 testing cases). In phantom validation, noise power spectra (NPS) were compared in water background and tissue‐mimicking inserts, using peak frequency and mean‐absolute‐error (MAE). In patient case evaluation, visual inspection and quantitative assessment were conducted on axial, coronal, and sagittal planes. Local and global noise texture were visually inspected in low‐dose CT images and the difference images between routine and low dose. Noise levels in liver and fat were measured. Local and global 2D Fourier magnitude spectra of the difference images and the corresponding radial mean profiles were used to assess similarity in noise frequency components within tissues and entire field‐of‐view, using spectral correlation mapper (SCM) and spectral angle mapper (SAM). Several baseline neural network models (e.g., GAN) were included in the evaluation. Statistical significance was tested using a t ‐test for related samples. Results PALETTE‐derived NPS showed accurate noise peak frequency (PALETTE/reference: water 1.40/1.40 lp/cm; inserts 1.7/1.7lp/cm) and small MAE (≤0.65 HU 2 cm 2 ). PALETTE created anatomy‐dependent noise texture, showing realistic local and global granularity and streaks. No statistically significant difference was observed in noise levels ( p > 0.05). Noise range was comparable across 3D image volume (PALETTE / reference):liver – [18.0, 53.4] / [19.3, 50.0] HU; fat – [11.7, 42.4] / [12.1, 41.3] HU. Percent absolute difference of local noise was small (mean ± standard deviation): liver 4.1%±3.1%, fat 4.6%±3.1%. Noise frequency distribution was close to the reference (mean per case): SCM ≥ 0.92, SAM ≤ 0.22. Additionally, PALETTE outperformed all baseline models in visual inspection and quantitative comparison. Conclusion PALETTE can provide high‐quality image domain noise insertion for simulating accurate low‐dose CT images created with a commercial non‐linear reconstruction algorithm.

Background SynthesiZed Improved Resolution and Concurrent nOise reductioN (ZIRCON) is a multi-kernel synthesis method that creates a single series of thin-slice computed tomography (CT) images displaying low noise and high spatial resolution, increasing reader efficiency and minimizing partial volume averaging. Purpose To compare the diagnostic performance of a single set of ZIRCON images to two routine clinical image series using conventional CT head and bone reconstruction kernels for diagnosing intracranial findings and fractures in patients with trauma or suspected acute neurologic deficit. Material and Methods In total, 50 patients underwent clinically indicated head CT in the ER (15 normal, 35 abnormal cases). A non-reader neuroradiologist established the reference standard. Three neuroradiologists reviewed two routine clinical series (head and bone kernels) and a single ZIRCON series, detecting intracranial findings or fractures and rating confidence (0–100). Sensitivity, specificity, and jackknife free-response receiver operating characteristic (JAFROC) figure of merit (FOM) were compared (limit of non-inferiority: −0.10). Results ZIRCON and conventional images demonstrated comparable performance for fractures (sensitivity: 51.5% vs. 54.5%; specificity: 40.2% vs. 34.2%) and intracranial findings (sensitivity: 88.2% vs. 91.4%; specificity: 77.2% vs. 73.7%).The estimated difference of JAFROC FOM demonstrated ZIRCON non-inferiority for acute pathologies overall (0.003 [95% CI=−0.051–0.057]) and fractures (0.048 [95% CI=−0.050–0.145]) but not for intracranial findings alone (−0.024 [95% CI=−0.100–0.052]). Conclusion Thin-slice, low noise, and high spatial resolution images can be created to display intracranial findings and fractures replacing multiple images series in head CT with similar performance. Future studies in more patients and further algorithmic development are warranted.

AbstractBackgroundThe first commercially available photon‐counting‐detector CT (PCD‐CT) has been introduced for clinical use. However, its spectral performance on single‐ and dual‐contrast imaging tasks has not been comprehensively assessed.PurposeTo evaluate the spectral imaging performance of a clinical PCD‐CT system for single‐contrast material [iodine (I) or gadolinium (Gd)] and dual‐contrast materials (I and Gd) in comparison with a dual‐source dual‐energy CT (DS‐DECT).MethodsIodine (5, 10, and 15 mg/mL) and gadolinium (3.3, 6.6, and 9.9 mg/mL) samples, and their mixtures (I/Gd: 5/3.3 and 10/6.6 mg/mL) were prepared and placed in two torso‐shaped water phantoms (lateral dimensions: 30 and 40 cm). These phantoms were scanned on a PCD‐CT (NAEOTOM Alpha, Siemens) at 90, 120, and 140 kV. The same phantoms were scanned on a DS‐DECT (SOMATOM Force, Siemens) with 70/Sn150, 80/Sn150, 90/Sn150, and 100/Sn150 kV. The radiation dose levels were matched [volume CT dose index (CTDIvol): 10 mGy for the 30 cm phantom and 20 mGy for the 40 cm phantom] across all tube voltage settings and between scanners. Two‐material decomposition (I/water or Gd/water) was performed on iodine or gadolinium samples, and three‐material decomposition (I/Gd/water) on both individual samples and mixtures. On each decomposed image, mean mass concentration (± standard deviation) was measured in circular region‐of‐interests placed on the contrast samples. Root‐mean‐square‐error (RMSE) values of iodine and gadolinium concentrations were reported based on the measurements across all contrast samples and repeated on 10 consecutive slices.ResultsFor all material decomposition tasks on the DS‐DECT, the kV pairs with greater spectral separation (70/Sn150 kV and 80/Sn150 kV) yielded lower RMSE values than other DS‐DECT and PCD‐CT alternatives. Specifically, for the optimal 70/Sn150 kV, RMSE values were 1.2 ± 0.1 mg/mL (I) for I/water material decomposition, 1.0 ± 0.1 mg/mL (Gd) for Gd/water material decomposition, and 4.5 ± 0.2 mg/mL (I) and 3.7 ± 0.2 mg/mL (Gd), respectively, for I/Gd/water material decomposition. On the PCD‐CT, the optimal tube voltages were 120 or 140 kV for I/water decomposition with RMSE values of 2.0 ± 0.1 mg/mL (I). For Gd/water decomposition on PCD‐CT, the optimal tube voltage was 140 kV with gadolinium RMSE values of 1.5 ± 0.1 mg/mL (Gd), with the 90 kV setting on PCD‐CT generating higher RMSE values for gadolinium concentration compared to all DS‐DECT and PCD‐CT alternatives. For three material decomposition, both imaging modalities demonstrated substantially higher RMSE values for iodine and gadolinium, with 90 kV being the optimal tube potential for Gd/I quantitation on PCD‐CT [5.4 ± 0.3 mg/mL (I) and 3.9 ± 0.2 mg/mL (Gd)], and DS‐DECT at 100/Sn150 kV having larger RMSE values for both materials compared to the alternatives for either modality.ConclusionOptimal tube voltage for material decomposition on the clinical PCD‐CT is task‐dependent but inferior to DS‐DECT using 70/Sn150 kV or 80/Sn150 kV in two‐material decomposition for single‐contrast imaging (iodine/water or gadolinium/water). Three material decomposition (iodine/gadolinium/water) in dual‐contrast imaging yields substantially higher RMSE for both imaging platforms.

AbstractPurposeA dual‐source CT system can be operated in a high‐pitch helical mode to provide a temporal resolution of 66 ms, which reduces motion artifacts in CT pulmonary angiography (CTPA). It can also be operated in a multi‐energy (ME) mode to provide iodine maps, beneficial in the evaluation of pulmonary embolism (PE). No energy‐integrating detector (EID) CT can perform simultaneous ME and high‐pitch acquisition. This phantom study aimed to evaluate the ability of a photon‐counting‐detector (PCD) CT to perform simultaneous high‐pitch and ME imaging for CTPA.MethodsA motion phantom was used to mimic the respiratory motion. Two tubes filled with iodine with intravascular thrombus mimicked by injecting glue within the tubes were placed along with 5, 10, and 15 mg/mL iodine samples, on a motion phantom at 20 and 30 revolutions per minute. Separate high‐pitch and ME EID‐CT scans and a single high‐pitch ME PCD scan were acquired and virtual monoenergetic images and iodine maps reconstructed. Percent thrombus occlusion was measured and compared between static and moving images.ResultsWhen there was motion, EID‐CT ME suffered from significant motion artifacts. The measured iodine concentrations with PCD‐CT in high‐pitch ME were more stable when there was a motion, with a lower standard deviation than EID‐CT in ME mode. The estimated percent thrombus occlusion dropped significantly with applied motion on EID‐CT, while PCD‐CT high‐pitch ME mode showed good agreement between measurements on static or moving images.ConclusionPCD‐CT with combined ME and high‐pitch mode facilitates simultaneous accurate iodine quantification and assessment of intravascular occlusion.