Artificial intelligence for differentiating COVID-19 from other viral pneumonias on CT: comparative analysis of different models based on quantitative and radiomic approaches.

BACKGROUND: To develop a pipeline for automatic extraction of quantitative metrics and radiomic features from lung computed tomography (CT) and develop artificial intelligence (AI) models supporting differential diagnosis between coronavirus disease 2019 (COVID-19) and other viral pneumonia (non-COVID-19). METHODS: Chest CT of 1,031 patients (811 for model building; 220 as independent validation set (IVS) with positive swab for severe acute respiratory syndrome coronavirus-2 (647 COVID-19) or other respiratory viruses (384 non-COVID-19) were segmented automatically. A Gaussian model, based on the HU histogram distribution describing well-aerated and ill portions, was optimised to calculate quantitative metrics (QM, n = 20) in both lungs (2L) and four geometrical subdivisions (GS) (upper front, lower front, upper dorsal, lower dorsal; n = 80). Radiomic features (RF) of first (RF1, n = 18) and second (RF2, n = 120) order were extracted from 2L using PyRadiomics tool. Extracted metrics were used to develop four multilayer-perceptron classifiers, built with different combinations of QM and RF: Model1 (RF1-2L); Model2 (QM-2L, QM-GS); Model3 (RF1-2L, RF2-2L); Model4 (RF1-2L, QM-2L, GS-2L, RF2-2L). RESULTS: The classifiers showed accuracy from 0.71 to 0.80 and area under the receiving operating characteristic curve (AUC) from 0.77 to 0.87 in differentiating COVID-19 versus non-COVID-19 pneumonia. Best results were associated with Model3 (AUC 0.867 ± 0.008) and Model4 (AUC 0.870 ± 0.011. For the IVS, the AUC values were 0.834 ± 0.008 for Model3 and 0.828 ± 0.011 for Model4. CONCLUSIONS: Four AI-based models for classifying patients as COVID-19 or non-COVID-19 viral pneumonia showed good diagnostic performances that could support clinical decisions.

A single phantom, a single statistical method for low-contrast detectability assessment.

PURPOSE: The assessment of low-contrast-details is a part of the quality control (QC) program in digital radiology. It generally consists of evaluating the threshold contrast (Cth) detectability details for different-sized inserts, appropriately located in dedicated QC test tools. This work aims to propose a simplified method, based on a statistical model approach for threshold contrast estimation, suitable for different modalities in digital radiology. METHODS: A home-madelow-contrast phantom, made of a central aluminium insert with a step-wedge, was assembled and tested. The reliability and robustness of the method were investigated for Mammography, Digital Radiography, Fluoroscopy and Angiography. Imageswere analysed using our dedicated software developed on Matlab®. TheCth is expressed in the same unit (mmAl) for all studied modalities. RESULTS: This method allows the collection of Cthinformation from different modalities and equipment by different vendors, and it could be used to define typical values. Results are summarized in detail. For 0.5 diameter detail, Cthresults are in the range of: 0.018-0.023 mmAl for 2D mammography and 0.26-0.34 mmAl DR images. For angiographic images, for 2.5 mm diameter detail, the Cths median values are 0.55, 0.4, 0.06, 0.12 mmAl for low dose fluoroscopy, coronary fluorography, cerebral and abdominal DSA, respectively. CONCLUSIONS: The statistical method proposed in this study gives a simple approach for Low-Contrast-Details assessment, and the typical values proposed can be implemented in a QA program for digital radiology modalities.

Correlation between lung ultrasound and chest CT patterns with estimation of pulmonary burden in COVID-19 patients.

PURPOSE: The capability of lung ultrasound (LUS) to distinguish the different pulmonary patterns of COVID-19 and quantify the disease burden compared to chest CT is still unclear. METHODS: PCR-confirmed COVID-19 patients who underwent both LUS and chest CT at the Emergency Department were retrospectively analysed. In both modalities, twelve peripheral lung zones were identified and given a Severity Score basing on main lesion pattern. On CT scans the well-aerated lung volume (%WALV) was visually estimated. Per-patient and per-zone assessments of LUS classification performance taking CT findings as reference were performed, further revisioning the images in case of discordant results. Correlations between number of disease-positive lung zones, Severity Score and %WALV on both LUS and CT were assessed. The area under receiver operating characteristic curve (AUC) was calculated to determine LUS performance in detecting %WALV ≤ 70 %. RESULTS: The study included 219 COVID-19 patients with abnormal chest CT. LUS correctly identified as positive 217 (99 %) patients, but per-zone analysis showed sensitivity = 75 % and specificity = 66 %. The revision of the 121 (55 %) cases with positive LUS and negative CT revealed COVID-compatible lesions in 42 (38 %) CT scans. Number of disease-positive zones, Severity Score and %WALV between LUS and CT showed moderate correlations. The AUCs for LUS Severity Score and number of LUS-positive zones did not differ in detecting %WALV ≤ 70 %. CONCLUSION: LUS in COVID-19 is valuable for case identification but shows only moderate correlation with CT findings as for lesion patterns and severity quantification. The number of disease-positive lung zones in LUS alone was sufficient to discriminate relevant disease burden.

Impact of inter-reader contouring variability on textural radiomics of colorectal liver metastases.

BACKGROUND: Radiomics is expected to improve the management of metastatic colorectal cancer (CRC). We aimed at evaluating the impact of liver lesion contouring as a source of variability on radiomic features (RFs). METHODS: After Ethics Committee approval, 70 liver metastases in 17 CRC patients were segmented on contrast-enhanced computed tomography scans by two residents and checked by experienced radiologists. RFs from grey level co-occurrence and run length matrices were extracted from three-dimensional (3D) regions of interest (ROIs) and the largest two-dimensional (2D) ROIs. Inter-reader variability was evaluated with Dice coefficient and Hausdorff distance, whilst its impact on RFs was assessed using mean relative change (MRC) and intraclass correlation coefficient (ICC). For the main lesion of each patient, one reader also segmented a circular ROI on the same image used for the 2D ROI. RESULTS: The best inter-reader contouring agreement was observed for 2D ROIs according to both Dice coefficient (median 0.85, interquartile range 0.78-0.89) and Hausdorff distance (0.21 mm, 0.14-0.31 mm). Comparing RF values, MRC ranged 0-752% for 2D and 0-1567% for 3D. For 24/32 RFs (75%), MRC was lower for 2D than for 3D. An ICC > 0.90 was observed for more RFs for 2D (53%) than for 3D (34%). Only 2/32 RFs (6%) showed a variability between 2D and circular ROIs higher than inter-reader variability. CONCLUSIONS: A 2D contouring approach may help mitigate overall inter-reader variability, albeit stable RFs can be extracted from both 3D and 2D segmentations of CRC liver metastases.

Radiomics predicts response of individual HER2-amplified colorectal cancer liver metastases in patients treated with HER2-targeted therapy.

The aim of our study was to develop and validate a machine learning algorithm to predict response of individual HER2-amplified colorectal cancer liver metastases (lmCRC) undergoing dual HER2-targeted therapy. Twenty-four radiomics features were extracted after 3D manual segmentation of 141 lmCRC on pretreatment portal CT scans of a cohort including 38 HER2-amplified patients; feature selection was then performed using genetic algorithms. lmCRC were classified as nonresponders (R-), if their largest diameter increased more than 10% at a CT scan performed after 3 months of treatment, responders (R+) otherwise. Sensitivity, specificity, negative (NPV) and positive (PPV) predictive values in correctly classifying individual lesion and overall patient response were assessed on a training dataset and then validated on a second dataset using a Gaussian naïve Bayesian classifier. Per-lesion sensitivity, specificity, NPV and PPV were 89%, 85%, 93%, 78% and 90%, 42%, 73%, 71% respectively in the testing and validation datasets. Per-patient sensitivity and specificity were 92% and 86%. Heterogeneous response was observed in 9 of 38 patients (24%). Five of nine patients were carriers of nonresponder lesions correctly classified as such by our radiomics signature, including four of seven harboring only one nonresponder lesion. The developed method has been proven effective in predicting behavior of individual metastases to targeted treatment in a cohort of HER2 amplified patients. The model accurately detects responder lesions and identifies nonresponder lesions in patients with heterogeneous response, potentially paving the way to multimodal treatment in selected patients. Further validation will be needed to confirm our findings.

Standardization of CT radiomics features for multi-center analysis: impact of software settings and parameters.

The aim of this multicentric study is an inter-center benchmarking, to assess how different set tools applied to the same radiomics workflow affected the radiomics features (RFs) values. This topic is of key importance to start collaboration between different centers and to bring radiomic studies from benchmark to bedside. A per-lesion analysis was performed on 56 metastases (mts) selected from 14 patients. A single radiologist performed the segmentation of all mts, and RFs were extracted from the same segmentation of each mts, using two different software and file formats. Potential sources of discrepancies were evaluated. The intraclass correlation coefficient was used to describe how strongly the same radiomic measurements calculated in the two different centers resemble each other. Moreover, means of the relative changes of each RF were calculated, compared and gradually reduced. We showed that, after matching all formulas, discrepancies in RFs calculation between two centers ranged from 1% to 277%. Therefore, we evaluated other sources of variability using a stepwise approach, which led us to reduce the inter-center discrepancies to 0% for 22/25 RFs and below 2% for 3 RFs out of 25. In this study we demonstrated that different radiomic applications and masks formats might strongly impact the computation of some RFs. Therefore, when dealing with multi-center studies it is mandatory to adopt all strategies that can help in limiting the differences, thus keeping in mind the feasibility of these strategies in large cohort studies.

Validation of a dose tracking software for skin dose map calculation in interventional radiology.

PURPOSE: Validate the skin dose software within the radiation dose index monitoring system NEXO[DOSE]® (Bracco Injeneering S.A., Lausanne, Switzerland). It provides the skin dose distribution in interventional radiology (IR) procedures. METHODS: To determine the skin dose distribution and the Peak Skin Dose (PSD) in IR procedures, the software uses exposure and geometrical parameters taken from the radiation dose structured report and additional information specific to each angiographic system. To test the accuracy of the software, GafChromic® XR-RV3 films, wrapped under a cylindrical PMMA phantom, were irradiated with different setups. Calculations and films results are compared in terms of absolute dose and geometric accuracy, using two angiographic systems (Philips Integris Allura FD20, Siemens AXIOM-ArtisZeego). RESULTS: Calculated and film measured PSD values agree with an average difference of 7% ± 5%. The discrepancies in dose evaluation increase up to 33% in lower dose regions, because the algorithm does not consider the out-of-field scatter contribution of the neighboring fields, which is more significant in these areas. Regarding the geometric accuracy, the differences between the simulated dose spatial distributions and the measured ones are<3 mm (4%) in simple tests and 5 mm (5%) in setups closer to clinical practice. Moreover, similar results are obtained for the two studied angiographic system vendors. CONCLUSIONS: NEXO[DOSE]® provides an accurate skin dose distribution and PSD estimate. It will allow faster and more accurate monitoring of patient follow-up in the future.

Patient organ and effective dose estimation in CT: comparison of four software applications.

BACKGROUND: Radiation dose in computed tomography (CT) has become a topic of high interest due to the increasing numbers of CT examinations performed worldwide. Hence, dose tracking and organ dose calculation software are increasingly used. We evaluated the organ dose variability associated with the use of different software applications or calculation methods. METHODS: We tested four commercial software applications on CT protocols actually in use in our hospital: CT-Expo, NCICT, NCICTX, and Virtual Dose. We compared dose coefficients, estimated organ doses and effective doses obtained by the four software applications by varying exposure parameters. Our results were also compared with estimates reported by the software authors. RESULTS: All four software applications showed dependence on tube voltage and volume CT dose index, while only CT-Expo was also dependent on other exposure parameters, in particular scanner model and pitch caused a variability till 50%. We found a disagreement between our results and those reported by the software authors (up to 600%), mainly due to a different extent of examined body regions. The relative range of the comparison of the four software applications was within 35% for most organs inside the scan region, but increased over the 100% for organs partially irradiated and outside the scan region. For effective doses, this variability was less evident (ranging from 9 to 36%). CONCLUSIONS: The two main sources of organ dose variability were the software application used and the scan region set. Dose estimate must be related to the process used for its calculation.

Sensitivity of CE-MRI in detecting local recurrence after radical prostatectomy.

AIM: The aim of our study was to evaluate the sensitivity of contrast-enhanced magnetic resonance (CE-MR) with phased array coil in the diagnosis of local recurrence in patients with prostate cancer after radical prostatectomy and referred for salvage radiotherapy (SRT). MATERIALS AND METHODS: This retrospective study included 73 patients treated with SRT after radical prostatectomy in the period between September 2006 and November 2017. All patients performed a CE-MRI with phased array coil before the start of SRT. A total of 213 patients treated at the ASST Grande Ospedale Metropolitano Niguarda in the period between September 2006 and November 2017 with SRT after radical prostatectomy were reviewed. Seventy-three patients with a CE-MRI with phased array coil of the pelvis before the start of SRT were included in the present study. RESULTS: At imaging review, recurrence local recurrent disease was diagnosed in 48 of 73 patients. By considering as reference standard the decrease in prostate-specific antigen (PSA) value after radiotherapy, we defined: 41 true positive (patients with MRI evidence of local recurrence and PSA value decreasing after SRT), 7 false positive (patients with MRI evidence of local recurrence without biochemical response after SRT), 3 true negative (patients without MRI evidence of local recurrence and stable or increased PSA value after SRT) and 22 false negative (patients without MRI evidence of local recurrence and PSA value decreasing after SRT) cases. The sensitivity values were calculated in relation to the PSA value before the start of treatment, obtaining a value of 74% for PSA above 0.2 ng/mL. CONCLUSION: The sensitivity of CE-MRI in local recurrence detection after radical prostatectomy increases with increasing PSA values. CE-MRI with phased array coil can detect local recurrences after radical prostatectomy with a good sensitivity in patients with pre-RT PSA value above 0.2 ng/mL.

Analysis of a multicentre cloud-based CT dosimetric database: preliminary results.

BACKGROUND: To manage and analyse dosimetric data provided by computed tomography (CT) scanners from four Italian hospitals. METHODS: A radiation dose index monitoring (RDIM) software was used to collect anonymised exams stored in a cloud server. Since hospitals use different names for the same procedure, digital imaging and communications in medicine (DICOM) tags more appropriate to describe exams were selected and associated to study common names (SCNs) from a radiology playbook according to scan region and use of contrast media. Retrospective analysis was carried out to describe population and to evaluate dosimetric indexes and inaccuracies associated with SCNs. RESULTS: More than 400 procedures were clustered into 95 SCNs, but 78% of exams on adults were described with only 10 SCNs. Median values of dose-length product (DLP) and volumetric CT dose index (CTDIvol) for three analysed SCNs were in agreement with those previously published. The percentage of inaccuracies does not heavily affect the dosimetric analysis on the whole cloud, since variations in median values reached at most 8%. CONCLUSIONS: Implementation of a cloud-based RDIM software and related issues were described, showing the strength of the chosen playbook-based clustering and its usefulness for homogeneous data analysis. This approach may allow for optimisation actions, accurate assessment of the risk associated with radiation exposure, comparison of different facilities, and, last but not least, collection of information for the implementation of the 2013/59 Euratom Directive.