Weakly-supervised learning-based pathology detection and localization in 3D chest CT scans.

BACKGROUND: Recent advancements in anomaly detection have paved the way for novel radiological reading assistance tools that support the identification of findings, aimed at saving time. The clinical adoption of such applications requires a low rate of false positives while maintaining high sensitivity. PURPOSE: In light of recent interest and development in multi pathology identification, we present a novel method, based on a recent contrastive self-supervised approach, for multiple chest-related abnormality identification including low lung density area ("LLDA"), consolidation ("CONS"), nodules ("NOD") and interstitial pattern ("IP"). Our approach alerts radiologists about abnormal regions within a computed tomography (CT) scan by providing 3D localization. METHODS: We introduce a new method for the classification and localization of multiple chest pathologies in 3D Chest CT scans. Our goal is to distinguish four common chest-related abnormalities: "LLDA", "CONS", "NOD", "IP" and "NORMAL". This method is based on a 3D patch-based classifier with a Resnet backbone encoder pretrained leveraging recent contrastive self supervised approach and a fine-tuned classification head. We leverage the SimCLR contrastive framework for pretraining on an unannotated dataset of randomly selected patches and we then fine-tune it on a labeled dataset. During inference, this classifier generates probability maps for each abnormality across the CT volume, which are aggregated to produce a multi-label patient-level prediction. We compare different training strategies, including random initialization, ImageNet weight initialization, frozen SimCLR pretrained weights and fine-tuned SimCLR pretrained weights. Each training strategy is evaluated on a validation set for hyperparameter selection and tested on a test set. Additionally, we explore the fine-tuned SimCLR pretrained classifier for 3D pathology localization and conduct qualitative evaluation. RESULTS: Validated on 111 chest scans for hyperparameter selection and subsequently tested on 251 chest scans with multi-abnormalities, our method achieves an AUROC of 0.931 (95% confidence interval [CI]: [0.9034, 0.9557], p $ p$ -value < 0.001) and 0.963 (95% CI: [0.952, 0.976], p $ p$ -value < 0.001) in the multi-label and binary (i.e., normal versus abnormal) settings, respectively. Notably, our method surpasses the area under the receiver operating characteristic (AUROC) threshold of 0.9 for two abnormalities: IP (0.974) and LLDA (0.952), while achieving values of 0.853 and 0.791 for NOD and CONS, respectively. Furthermore, our results highlight the superiority of incorporating contrastive pretraining within the patch classifier, outperforming Imagenet pretraining weights and non-pretrained counterparts with uninitialized weights (F1 score = 0.943, 0.792, and 0.677 respectively). Qualitatively, the method achieved a satisfactory 88.8% completeness rate in localization and maintained an 88.3% accuracy rate against false positives. CONCLUSIONS: The proposed method integrates self-supervised learning algorithms for pretraining, utilizes a patch-based approach for 3D pathology localization and develops an aggregation method for multi-label prediction at patient-level. It shows promise in efficiently detecting and localizing multiple anomalies within a single scan.

Ultra-low dose chest CT for the diagnosis of pulmonary arteriovenous malformation in patients with hereditary hemorrhagic telangiectasia.

PURPOSE: The purpose of this study was to compare ultra-low dose (ULD) and standard low-dose (SLD) chest computed tomography (CT) in terms of radiation exposure, image quality and diagnostic value for diagnosing pulmonary arteriovenous malformation (AVM) in patients with hereditary hemorrhagic telangiectasia (HHT). MATERIALS AND METHODS: In this prospective board-approved study consecutive patients with HHT referred to a reference center for screening and/or follow-up chest CT examination were prospectively included from December 2020 to January 2022. Patients underwent two consecutive non-contrast chest CTs without dose modulation (i.e., one ULD protocol [80 kVp or 100 kVp, CTDIvol of 0.3 mGy or 0.6 mGy] and one SLD protocol [140 kVp, CTDIvol of 1.3 mGy]). Objective image noises measured at the level of tracheal carina were compared between the two protocols. Overall image quality and diagnostic confidence were scored on a 4-point Likert scale (1 = insufficient to 4 = excellent). Sensitivity, specificity, positive predictive value and negative predictive value of ULD CT for diagnosing pulmonary AVM with a feeding artery of over 2 mm in diameter were calculated along with their 95% confidence intervals (CI) using SLD images as the standard of reference. RESULTS: A total of 44 consecutive patients with HHT (31 women; mean age, 42 ± 16 [standard deviation (SD)] years; body mass index, 23.2 ± 4.5 [SD] kg/m2) were included. Thirty-four pulmonary AVMs with a feeding artery of over 2 mm in diameter were found with SLD images versus 35 with ULD images. Sensitivity, specificity, predictive positive value, and predictive negative value of ULD CT for the diagnosis of PAVM were 100% (34/34; 95% CI: 90-100), 96% (18/19; 95% CI: 74-100), 97% (34/35; 95% CI: 85-100) and 100% (18/18; 95% CI: 81-100), respectively. A significant difference in diagnostic confidence scores was found between ULD (3.8 ± 0.4 [SD]) and SLD (3.9 ± 0.1 [SD]) CT images (P = 0.03). No differences in overall image quality scores were found between ULD CT examinations (3.9 ± 0.2 [SD]) and SLD (4 ± 0 [SD]) CT examinations (P = 0.77). Effective radiation dose decreased significantly by 78.8% with ULD protocol, with no significant differences in noise values between ULD CT images (16.7 ± 5.0 [SD] HU) and SLD images (17.7 ± 6.6 [SD] HU) (P = 0.07). CONCLUSION: ULD chest CT provides 100% sensitivity and 96% specificity for the diagnosis of treatable pulmonary AVM with a feeding artery of over 2 mm in diameter, leading to a 78.8% dose-saving compared with a standard low-dose protocol.

Bicolor K-edge spectral photon-counting CT imaging for the diagnosis of thoracic endoleaks: A dynamic phantom study.

PURPOSE: The purpose of this study was to investigate the feasibility of identifying and characterizing the three most common types of endoleaks within a thoracic aorta aneurysm model using bicolor K-edge imaging with a spectral photon-counting computing tomography (SPCCT) system in combination with a biphasic contrast agent injection. MATERIALS AND METHODS: Three types of thoracic endoleaks (type 1, 2 and 3) were created in a dynamic anthropomorphic thoracic aorta phantom. Protocol consisted in an injection of an iodinated contrast material followed 80 seconds after an injection of a gadolinium-based contrast agent (GBCA). The phantom was scanned using a clinical prototype SPCCT during bicolor phase imaging consisting in an early distribution of GBCA and a late distribution of iodine. Conventional and spectral images were reconstructed for differentiating between the contrast agents and measuring their respective attenuation values and concentrations inside and outside the stent graft. RESULTS: Conventional images failed to provide specific dynamic imaging contrast agents in the aneurysmal sac and outside the stent graft while spectral images differentiated their specific distribution. In type 1 and 3 thoracic endoleaks, GBCA concentration was measured outside the stent graft at 6.1 ± 3.7 (standard deviation [SD]) mg/mL and 6.0 ± 4.0 (SD) mg/mL, respectively, in favor of an early blood flow. In type 2 thoracic endoleak, iodine was measured outside the stent graft at 24.3 ± 5.5 (SD) mg/mL in favor of a late blood flow in the aneurysmal sac. CONCLUSION: Bicolor K-edge imaging enabled SPCCT allows a bicolor characterization of thoracic aorta endoleaks in a single acquisition in combination with a biphasic contrast agent injection.

High-resolution synchrotron K-edge subtraction CT allows tracking and quantifying therapeutic cells and their scaffold in a rat model of focal cerebral injury and can serve as a reference for spectral photon counting CT.

Background: The objective of this study was to demonstrate that synchrotron K-edge subtraction tomography (SKES-CT) can simultaneously track therapeutic cells and their encapsulating carrier, in vivo, in a rat model of focal brain injury using a dual-contrast agent approach. The second objective was to determine if SKES-CT could be used as a reference method for spectral photon counting tomography (SPCCT). Methods: Phantoms containing different concentrations of gold and iodine nanoparticles (AuNPS/INPs) were imaged with SKES-CT and SPCCT to assess their performances. A pre-clinical study was performed in rats with focal cerebral injury which intracerebrally received AuNPs-labelled therapeutic cells encapsulated in a INPs-labelled scaffold. Animals were imaged in vivo with SKES-CT and back-to-back with SPCCT. Results: SKES-CT revealed to be reliable for quantification of gold and iodine, whether alone or mixed. In the preclinical model, SKES-CT showed that AuNPs remained at the site of cell injection, while INPs expanded within and/or along the lesion border, suggesting dissociation of both components in the first days post-administration. Compared to SKES-CT, SPCCT was able to correctly locate gold, but not completely located iodine. When SKES-CT was used as reference, SPCCT gold quantification appeared very accurate both in vitro and in vivo. Iodine quantification by SPCCT was also quite accurate, albeit less so than for gold. Conclusion: We here provide the proof-of-concept that SKES-CT is a novel method of choice for performing dual-contrast agent imaging in the context of brain regenerative therapy. SKES-CT may also serve as ground truth for emerging technologies such as multicolour clinical SPCCT.

Additional value of chest CT AI-based quantification of lung involvement in predicting death and ICU admission for COVID-19 patients.

Objectives: We evaluated the contribution of lung lesion quantification on chest CT using a clinical Artificial Intelligence (AI) software in predicting death and intensive care units (ICU) admission for COVID-19 patients. Methods: For 349 patients with positive COVID-19-PCR test that underwent a chest CT scan at admittance or during hospitalization, we applied the AI for lung and lung lesion segmentation to obtain lesion volume (LV), and LV/Total Lung Volume (TLV) ratio. ROC analysis was used to extract the best CT criterion in predicting death and ICU admission. Two prognostic models using multivariate logistic regressions were constructed to predict each outcome and were compared using AUC values. The first model ("Clinical") was based on patients' characteristics and clinical symptoms only. The second model ("Clinical+LV/TLV") included also the best CT criterion. Results: LV/TLV ratio demonstrated best performance for both outcomes; AUC of 67.8% (95% CI: 59.5 - 76.1) and 81.1% (95% CI: 75.7 - 86.5) respectively. Regarding death prediction, AUC values were 76.2% (95% CI: 69.9 - 82.6) and 79.9% (95%IC: 74.4 - 85.5) for the "Clinical" and the "Clinical+LV/TLV" models respectively, showing significant performance increase (+ 3.7%; p-value<0.001) when adding LV/TLV ratio. Similarly, for ICU admission prediction, AUC values were 74.9% (IC 95%: 69.2 - 80.6) and 84.8% (IC 95%: 80.4 - 89.2) respectively corresponding to significant performance increase (+ 10%: p-value<0.001). Conclusions: Using a clinical AI software to quantify the COVID-19 lung involvement on chest CT, combined with clinical variables, allows better prediction of death and ICU admission.

Spectral Photon-Counting CT Technology in Chest Imaging.

The X-ray imaging field is currently undergoing a period of rapid technological innovation in diagnostic imaging equipment. An important recent development is the advent of new X-ray detectors, i.e., photon-counting detectors (PCD), which have been introduced in recent clinical prototype systems, called PCD computed tomography (PCD-CT) or photon-counting CT (PCCT) or spectral photon-counting CT (SPCCT) systems. PCD allows a pixel up to 200 microns pixels at iso-center, which is much smaller than that can be obtained with conventional energy integrating detectors (EID). PCDs have also a higher dose efficiency than EID mainly because of electronic noise suppression. In addition, the energy-resolving capabilities of these detectors allow generating spectral basis imaging, such as the mono-energetic images or the water/iodine material images as well as the K-edge imaging of a contrast agent based on atoms of high atomic number. In recent years, studies have therefore been conducted to determine the potential of PCD-CT as an alternative to conventional CT for chest imaging.

Diagnostic Performance of Extracellular Volume Quantified by Dual-Layer Dual-Energy CT for Detection of Acute Myocarditis.

BACKGROUND: Myocardial extracellular volume (ECV) is a marker of the myocarditis inflammation burden and can be used for acute myocarditis diagnosis. Dual-energy computed tomography (DECT) enables its quantification with high concordance with cardiac magnetic resonance (CMR). PURPOSE: To investigate the diagnostic performance of myocardial ECV quantified on a cardiac dual-layer DECT in a population of patients with suspected myocarditis, in comparison to CMR. METHODS: 78 patients were included in this retrospective monocenter study, 60 were diagnosed with acute myocarditis and 18 patients were considered as a control population, based on the 2009 Lake and Louise criteria. All subjects underwent a cardiac DECT in acute phase consisted in an arterial phase followed by a late iodine enhancement phase at 10 min after injection (1.2 mL/kg, iodinated contrast agent). ECV was calculated using the hematocrit level measured the day of DECT examinations. Non-parametric analyses have been used to test the differences between groups and the correlations between the variables. A ROC curve has been used to identify the optimal ECV cut-off discriminating value allowing the detection of acute myocarditis cases. A p value < 0.05 has been considered as significant. RESULTS: The mean ECV was significantly higher (p < 0.001) for the myocarditis group compared to the control (34.18 ± 0.43 vs. 30.04 ± 0.53%). A cut-off value of ECV = 31.60% (ROC AUC = 0.835, p < 0.001) allows to discriminate the myocarditis with a sensitivity of 80% and a specificity of 78% (positive predictive value = 92.3%, negative predictive value = 53.8% and accuracy = 79.5%). CONCLUSION: Myocardial ECV enabled by DECT allows to diagnose the acute myocarditis with a cut-off at 31.60% for a sensitivity of 80% and specificity of 78%.