Liver lesion localisation and classification with convolutional neural networks: a comparison between conventional and spectral computed tomography.

PURPOSE: To evaluate the benefit of the additional available information present in spectral CT datasets, as compared to conventional CT datasets, when utilizing convolutional neural networks for fully automatic localisation and classification of liver lesions in CT images. MATERIALS AND METHODS: Conventional and spectral CT images (iodine maps, virtual monochromatic images (VMI)) were obtained from a spectral dual-layer CT system. Patient diagnosis were known from the clinical reports and classified into healthy, cyst and hypodense metastasis. In order to compare the value of spectral versus conventional datasets when being passed as input to machine learning algorithms, we implemented a weakly-supervised convolutional neural network (CNN) that learns liver lesion localisation without pixel-level ground truth annotations. Regions-of-interest are selected automatically based on the localisation results and are used to train a second CNN for liver lesion classification (healthy, cyst, hypodense metastasis). The accuracy of lesion localisation was evaluated using the Euclidian distances between the ground truth centres of mass and the predicted centres of mass. Lesion classification was evaluated by precision, recall, accuracy and F1-Score. RESULTS: Lesion localisation showed the best results for spectral information with distances of 8.22 ± 10.72 mm, 8.78 ± 15.21 mm and 8.29 ± 12.97 mm for iodine maps, 40 keV and 70 keV VMIs, respectively. With conventional data distances of 10.58 ± 17.65 mm were measured. For lesion classification, the 40 keV VMIs achieved the highest overall accuracy of 0.899 compared to 0.854 for conventional data. CONCLUSION: An enhanced localisation and classification is reported for spectral CT data, which demonstrates that combining machine-learning technology with spectral CT information may in the future improve the clinical workflow as well as the diagnostic accuracy.

Sparse sampling computed tomography (SpSCT) for detection of pulmonary embolism: a feasibility study.

OBJECTIVES: Evaluation of sparse sampling computed tomography (SpSCT) regarding subjective and objective image criteria for the detection of pulmonary embolism (PE) at different simulated dose levels. METHODS: Computed tomography pulmonary angiography (CTPA) scans of 20 clinical patients were used to obtain simulated low-dose scans with 100%-50%-25%-12.5%-6.3%-3.1% of the clinical dose, resulting in a total of six dose levels (DL). From these full sampling (FS) data, every second (2-SpSCT) or fourth (4-SpSCT) projection was used to obtain simulated sparse sampling scans. Each image set was evaluated by four blinded radiologists regarding subjective image criteria (artifacts, image quality) and diagnostic performance (confidence, sensitivity, specificity, accuracy, and area under the curve). Additionally, the contrast-to-noise ratio (CNR) was evaluated for objective image quality. RESULTS: Sensitivity was 100% with 2-SpSCT and 4-SpSCT at the 25% DL and the 12.5% DL for all localizations of PE (one subgroup 98.5%). With FS, the sensitivity decreased to 90% at the 12.5% DL. 2-SpSCT and 4-SpSCT showed higher values for sensitivity, specificity, accuracy, and the area under the curve at all DL compared with FS. Subjective image quality was significantly higher for 4-SpSCT compared with FS at each dose level (p < 0.01, paired t test). Only with 4-SpSCT, all examinations were rated as showing diagnostic image quality at the 12.5% DL. CONCLUSIONS: Via SpSCT, a dose reduction down to a 12.5% dose level (corresponding to a mean effective dose of 0.38 mSv in the current study) for CTPA is possible while maintaining high image quality and full diagnostic confidence. KEY POINTS: • With sparse sampling CT, radiation dose could be significantly reduced in clinical routine. • Sparse sampling CT is a novel hardware solution with which less projection images are acquired. • In the current study, a dose reduction of 87.5% (corresponding to a mean effective dose of 0.38 mSv) for CTPA could be achieved while maintaining excellent diagnostic performance.

Differentiating intrapulmonary metastases from different primary tumors via quantitative dual-energy CT based iodine concentration and conventional CT attenuation.

INTRODUCTION: To investigate the utility of quantitative dual-energy spectral CT derived iodine concentration (IC), in comparison with conventional CT attenuation, for the differentiation of pulmonary metastases from different primary malignancies. MATERIALS AND METHODS: CT scans were performed on a dual-layer spectral CT. We retrospectively evaluated pulmonary metastases of 130 patients (77 men and 53 women, mean age 63, range 22-87) with primary bone (OS) (osteosarcoma; n = 6), breast (invasive-ductal adenocarcinoma; n = 17), colorectal (CRC) (adenocarcinoma; n = 27), head and neck (HNC) (squamous cell carcinoma; n = 17), kidney (RCC) (clear-cell renal cell carcinoma; n = 10), lung (adenocarcinoma; n = 12), pancreato-biliary (PBC) (adenocarcinoma; n = 18), prostate (adenocarcinoma; n = 5), soft tissue (undifferentiated pleomorphic sarcoma; n = 6), skin (malignant melanoma; n = 6), and urinary tract (transitional-cell carcinoma; n = 6) malignancies. Quantitative IC and conventional CT numbers (HU) were extracted and normalized to the thoracic aorta. Differences between the groups were assessed by pairwise t-tests with Holm-Sidak post-hoc p-value adjustment for multiple comparisons. Diagnostic accuracy was evaluated by receiver operating characteristic (ROC) analysis. RESULTS: Significant differences in IC and HU were noted for pulmonary metastases from RCC (IC: 2.83 mg/ml; HU: 93.12) versus breast cancer (IC: 1.47 mg/ml, adjusted p < 0.05; HU: 59.57, adjusted p < 0.05), CRC (IC: 1.23 mg/ml, adjusted p < 0.001; HU: 49.82, adjusted p < 0.001) and HNC (IC: 1.54 mg/ml, adjusted p < 0.05; HU: 58.91, adjusted p < 0.01). Based on IC alone, significant differences were further observed between metastatic lesions from CRC versus OS (IC: 2.36 mg/ml, adjusted p < 0.001), PBC (IC: 2.16 mg/ml, adjusted p < 0.001) and urinary tract carcinoma (IC: 2.21 mg/ml, adjusted p < 0.05). Based on IC and HU, pulmonary metastases from OS, HNC and RCC may be differentiated from other pulmonary metastases (area under ROC curve, 0.69-0.79). The diagnostic accuracy to discriminate between pulmonary metastases from PBC and those from other malignancies was significantly higher based on IC as compared to HU (area under ROC curve, 0.66; p < 0.05); no significant differences in diagnostic accuracy were noted for other differentiations. CONCLUSIONS: Our findings demonstrate the utility of both dual-energy CT derived quantitative IC and conventional CT attenuation values for the differential diagnosis in suspected pulmonary metastases of unknown origin, however giving preference to the use of IC.

Evaluation of a computed-tomography-based assessment scheme in treatment decision-making for isolated orbital floor fractures.

INTRODUCTION: Treatment decisions for fractures of the orbital floor are based on clinical appearance, ophthalmological examination, and computed tomography (CT) scans. In extensive fractures, decisions are easily made between conservative and surgical treatment. However, objective parameters are rare in inconclusive cases. MATERIALS AND METHODS: Our retrospective study included 106 patients with unilateral isolated orbital floor fractures. Correlations between preoperative ophthalmological examinations and specific CT parameters were performed. RESULTS: The defect size of the fracture appeared to be significantly associated with the presence of diplopia. CT-morphological parameters and preoperative ophthalmological results showed statistical significance for diplopia and incarceration of inferior rectus muscle (IRM), diplopia and displacement of IRM, decreased mobility and incarceration of IRM, and decreased mobility and displacement of IRM. DISCUSSION: Our clinical assessment scheme for CT scans of orbital floor fractures is aimed at facilitating treatment decision making using four CT-based variables. As critical size defects of the orbital floor of ≥2 cm2 are likely to cause clinically significant posterior displacement of the globe, resulting in enophthalmos, the proposed parameters offer a readily accessible and easy to evaluate scheme that helps to identify patients in need of surgical intervention.

Dual layer computed tomography: Reduction of metal artefacts from posterior spinal fusion using virtual monoenergetic imaging.

INTRODUCTION: To evaluate the clinical potential of dual layer computed tomography (DLCT) for posterior fusions of the thoracic and lumbar spine and determine the optimal keV-settings for an improved overall image quality and effective reduction of metal artefacts affecting the implant inheriting vertebral body, the spinal canal, the paravertebral muscle and aorta. METHODS AND MATERIALS: Twenty patients with posterior thoracic and lumbar spinal fusion, who underwent a 120kVp- DLCT scan were included in this study. Two independent readers evaluated axial 0.9 mm slides with soft tissue and bone window settings. Image quality of the conventional scan was compared to virtual monoenergetic images (VMI) at 40, 60, 80, 100,120, 140, 160, 180 and 200 keV. Diagnostic image quality was assessed on a four point Likert-scale overall, as well as specifically for the implant inheriting bone, paravertebral muscle, spinal canal or aorta. The Hounsfield Units (HU) of the area with the most pronounced streak artefact as well as HU of a reference area containing fat and muscle were documented for each keV-setting and compared to the conventional image. SNR and CNR were calculated for each of the four anatomic areas. Statistical analysis was conducted for the total collective and separately for the thoracic and lumbar spine level. RESULTS: Starting from 80 keV qualitative analysis revealed significant improvement of overall image quality and benefit for each tissue separately compared to the conventional images (CI) (p-values in the range from <0.001 to 0.005). 180 keV was considered the optimal monoenergetic setting regarding the overall image quality. For the assessment of the implant inheriting bone, the spinal canal, paravertebral muscle and aorta 200, 180, 160 and 180 keV were rated to be the most sufficient. Our results reveal high inter-reader agreement for qualitative evaluations (intra-class correlation coefficients >0.927; p < 0.05). HU values within the most pronounced streak artefact increased significantly with higher keV (p < 0.001), while there was no significant alteration of HU within the reference area. A decrease in SNR and CNR for higher VMI was revealed by our results. CONCLUSION: VMIs of higher energies provide significant reduction of metallic artefacts from posterior spinal fusions. Dedicated keV settings to evaluate either the implant inheriting bone, the spinal canal,adjacent muscle or aorta - structures, which are frequently of particular interest after posterior spinal fusion - are recommended. In addition, an optimal keV for an improved overall image quality is proposed.

Dual-layer spectral computed tomography: Virtual non-contrast in comparison to true non-contrast images.

PURPOSE: To evaluate virtual-non-contrast (VNC) images obtained from clinical triphasic scans with a dual-layer spectral computed tomography system regarding accuracy of iodine subtraction. MATERIAL AND METHODS: From September to December 2016, 62 consecutive patients who underwent a clinical routine triphasic CT examination were included into this retrospective study. VNC images based on the arterial and portal venous phase were generated. For every patient and every contrast phase, a region-of-interest (ROI) was defined in aorta, liver, renal cortex, spongious bone, fat, muscle and fluid (i.e. gallbladder, urinary bladder), resulting in 2170 ROIs. VNC images were compared to true-non-contrast (TNC) images regarding difference in attenuation. Consistency between VNC images obtained from the arterial and portal venous phase as well as the influence of the initial attenuation on respective VNC images were evaluated. RESULTS: Comparison of HU in VNC and TNC images showed a high accuracy of iodine elimination. Mean difference between TNC and VNC images was only 0.5 ±â€¯8.5 HU and >90% of all comparisons showed a difference of less than 15 HU. For all tissues but spongious bone, mean absolute difference between TNC and VNC images was below 10 HU. VNC images derived from the arterial and the portal venous phase showed excellent correlation. The quality of iodine removal in VNC images was not influenced by the original contrast enhancement. However, VNC images cannot be used for evaluation of iodine removal in bone as bone and iodine can hardly be differentiated via spectral CT. CONCLUSION: VNC imaging in DL-CT is a promising tool for daily clinical routine. As non-enhanced CT images are essential in multiple clinical situations, the permanent availability of VNC images with dual-layer spectral CT will result in a substantial reduction of radiation exposure and an increased diagnostic value of monophasic contrast-enhanced CT scans.

Accuracy of iodine quantification in dual-layer spectral CT: Influence of iterative reconstruction, patient habitus and tube parameters.

PURPOSE: Evaluation of the influence of iterative reconstruction, tube settings and patient habitus on the accuracy of iodine quantification with dual-layer spectral CT (DL-CT). MATERIAL AND METHODS: A CT abdomen phantom with different extension rings and four iodine inserts (1, 2, 5 and 10 mg/ml) was scanned on a DL-CT. The phantom was scanned with tube-voltages of 120 and 140 kVp and CTDIvol of 2.5, 5, 10 and 20 mGy. Reconstructions were performed for eight levels of iterative reconstruction (i0-i7). Diagnostic dose levels are classified depending on patient-size and radiation dose. RESULTS: Measurements of iodine concentration showed accurate and reliable results. Taking all CTDIvol-levels into account, the mean absolute percentage difference (MAPD) showed less accuracy for low CTDIvol-levels (2.5 mGy: 34.72%) than for high CTDIvol-levels (20 mGy: 5.89%). At diagnostic dose levels, accurate quantification of iodine was possible (MAPD 3.38%). Level of iterative reconstruction did not significantly influence iodine measurements. Iodine quantification worked more accurately at a tube voltage of 140 kVp. Phantom size had a considerable effect only at low-dose-levels; at diagnostic dose levels the effect of phantom size decreased (MAPD <5% for all phantom sizes). CONCLUSION: With DL-CT, even low iodine concentrations can be accurately quantified. Accuracies are higher when diagnostic radiation doses are employed.

Experimental feasibility of spectral photon-counting computed tomography with two contrast agents for the detection of endoleaks following endovascular aortic repair.

OBJECTIVES: After endovascular aortic repair (EVAR), discrimination of endoleaks and intra-aneurysmatic calcifications within the aneurysm often requires multiphase computed tomography (CT). Spectral photon-counting CT (SPCCT) in combination with a two-contrast agent injection protocol may provide reliable detection of endoleaks with a single CT acquisition. METHODS: To evaluate the feasibility of SPCCT, the stent-lined compartment of an abdominal aortic aneurysm phantom was filled with a mixture of iodine and gadolinium mimicking enhanced blood. To represent endoleaks of different flow rates, the adjacent compartments contained either one of the contrast agents or calcium chloride to mimic intra-aneurysmatic calcifications. After data acquisition with a SPCCT prototype scanner with multi-energy bins, material decomposition was performed to generate iodine, gadolinium and calcium maps. RESULTS: In a conventional CT slice, Hounsfield units (HU) of the compartments were similar ranging from 147 to 168 HU. Material-specific maps differentiate the distributions within the compartments filled with iodine, gadolinium or calcium. CONCLUSION: SPCCT may replace multiphase CT to detect endoleaks without sacrificing diagnostic accuracy. It is a unique feature of our method to capture endoleak dynamics and allow reliable distinction from intra-aneurysmatic calcifications in a single scan, thereby enabling a significant reduction of radiation exposure. KEY POINTS: • SPCCT might enable advanced endoleak detection. • Material maps derived from SPCCT can differentiate iodine, gadolinium and calcium. • SPCCT may potentially reduce radiation burden for EVAR patients under post-interventional surveillance.

Assessment of quantification accuracy and image quality of a full-body dual-layer spectral CT system.

The performance of a recently introduced spectral computed tomography system based on a dual-layer detector has been investigated. A semi-anthropomorphic abdomen phantom for CT performance evaluation was imaged on the dual-layer spectral CT at different radiation exposure levels (CTDIvol of 10 mGy, 20 mGy and 30 mGy). The phantom was equipped with specific low-contrast and tissue-equivalent inserts including water-, adipose-, muscle-, liver-, bone-like materials and a variation in iodine concentrations. Additionally, the phantom size was varied using different extension rings to simulate different patient sizes. Contrast-to-noise (CNR) ratio over the range of available virtual mono-energetic images (VMI) and the quantitative accuracy of VMI Hounsfield Units (HU), effective-Z maps and iodine concentrations have been evaluated. Central and peripheral locations in the field-of-view have been examined. For all evaluated imaging tasks the results are within the calculated theoretical range of the tissue-equivalent inserts. Especially at low energies, the CNR in VMIs could be boosted by up to 330% with respect to conventional images using iDose/spectral reconstructions at level 0. The mean bias found in effective-Z maps and iodine concentrations averaged over all exposure levels and phantom sizes was 1.9% (eff. Z) and 3.4% (iodine). Only small variations were observed with increasing phantom size (+3%) while the bias was nearly independent of the exposure level (±0.2%). Therefore, dual-layer detector based CT offers high quantitative accuracy of spectral images over the complete field-of-view without any compromise in radiation dose or diagnostic image quality.