Artificial intelligence diagnostic accuracy in fracture detection from plain radiographs and comparing it with clinicians: a systematic review and meta-analysis

PURPOSE

Fracture detection is one of the most commonly used and studied aspects of artificial intelligence (AI) in medicine. In this systematic review and meta-analysis, we aimed to summarize available literature and data regarding AI performance in fracture detection on plain radiographs and various factors affecting it.

METHODS

We systematically reviewed studies evaluating AI algorithms in detecting bone fractures in plain radiographs, combined their performance using meta-analysis (a bivariate regression approach), and compared it with that of clinicians. We also analyzed the factors potentially affecting algorithm performance using meta-regression.

RESULTS

Our analysis included 100 studies. In 83 studies with confusion matrices, AI algorithms showed a sensitivity of 91.43% and a specificity of 92.12% (Area under the summary receiver operator curve = 0.968). After adjustment and false discovery rate correction, tibia/fibula (excluding ankle) fractures were associated with higher (7.0%, p=0.004) AI sensitivity, while more recent publications (5.5%, p=0.003) and Xception architecture (6.6%, p<0.001) were associated with higher specificity. Clinicians and AI showed similar specificity in fracture identification, although AI leaned to higher sensitivity (7.6%, p=0.07). Radiologists, on the other hand, were more specific than AI overall and in several subgroups, and more sensitive to hip fractures before FDR correction.

CONCLUSIONS

Currently available AI aids could result in a significant improvement in care where radiologists are not readily available. Moreover, identifying factors affecting algorithm performance could guide AI development teams in their process of optimizing their products.

Potential of photon counting computed tomography derived spectral reconstructions to reduce beam-hardening artifacts in chest CT

PURPOSE

Aim of the recent study is to point out a method to optimize quality of CT scans in oncological patients with port systems. This study investigates the potential of photon counting computed tomography (PCCT) for reduction of beam hardening artifacts caused by port-implants in chest imaging by means of spectral reconstructions.

METHOD

In this retrospective single-center study, 8 ROIs for 19 spectral reconstructions (polyenergetic imaging, monoenergetic reconstructions from 40 to 190 keV as well as iodine maps and virtual non contrast (VNC)) of 49 patients with pectoral port systems undergoing PCCT of the chest for staging of oncologic disease were measured. Mean values and standard deviation (SD) Hounsfield unit measurements of port-chamber associated hypo- and hyperdense artifacts, bilateral muscles and vessels has been carried out. Also, a structured assessment of artifacts and imaging findings was performed by two radiologists.

RESULTS

A significant association of keV with iodine contrast as well as artifact intensity was noted (all p < 0.001). In qualitative assessment, utilization of 120 keV monoenergetic reconstructions could reduce severe and pronounced artifacts completely, as compared to lower keV reconstructions (p < 0.001). Regarding imaging findings, no significant difference between monoenergetic reconstructions was noted (all p > 0.05). In cases with very high iodine concentrations in the subclavian vein, image distortions were noted at 40 keV images (p < 0.01).

CONCLUSIONS

The present study demonstrates that PCCT derived spectral reconstructions can be used in oncological imaging of the thorax to reduce port-derived beam-hardening artefacts. When evaluating image data sets within a staging, it can be particularly helpful to consider the 120 keV VMIs, in which the artefacts are comparatively low.

GeneAI 3.0: powerful, novel, generalized hybrid and ensemble deep learning frameworks for miRNA species classification of stationary patterns from nucleotides

Due to the intricate relationship between the small non-coding ribonucleic acid (miRNA) sequences, the classification of miRNA species, namely Human, Gorilla, Rat, and Mouse is challenging. Previous methods are not robust and accurate. In this study, we present AtheroPoint's GeneAI 3.0, a powerful, novel, and generalized method for extracting features from the fixed patterns of purines and pyrimidines in each miRNA sequence in ensemble paradigms in machine learning (EML) and convolutional neural network (CNN)-based deep learning (EDL) frameworks. GeneAI 3.0 utilized five conventional (Entropy, Dissimilarity, Energy, Homogeneity, and Contrast), and three contemporary (Shannon entropy, Hurst exponent, Fractal dimension) features, to generate a composite feature set from given miRNA sequences which were then passed into our ML and DL classification framework. A set of 11 new classifiers was designed consisting of 5 EML and 6 EDL for binary/multiclass classification. It was benchmarked against 9 solo ML (SML), 6 solo DL (SDL), 12 hybrid DL (HDL) models, resulting in a total of 11 + 27 = 38 models were designed. Four hypotheses were formulated and validated using explainable AI (XAI) as well as reliability/statistical tests. The order of the mean performance using accuracy (ACC)/area-under-the-curve (AUC) of the 24 DL classifiers was: EDL > HDL > SDL. The mean performance of EDL models with CNN layers was superior to that without CNN layers by 0.73%/0.92%. Mean performance of EML models was superior to SML models with improvements of ACC/AUC by 6.24%/6.46%. EDL models performed significantly better than EML models, with a mean increase in ACC/AUC of 7.09%/6.96%. The GeneAI 3.0 tool produced expected XAI feature plots, and the statistical tests showed significant p-values. Ensemble models with composite features are highly effective and generalized models for effectively classifying miRNA sequences.

Setting up regional diagnostic reference levels for pediatric computed tomography in Latin America: preliminary results, challenges and the work ahead

We established a framework for collecting radiation doses for head, chest and abdomen-pelvis computed tomography (CT) in children scanned at multiple imaging sites across Latin America with an aim towards establishing diagnostic reference levels (DRLs) and achievable doses (ADs) in pediatric CT in Latin America. Our study included 12 Latin American sites (in Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Honduras and Panama) contributing data on the four most common pediatric CT examinations (non-contrast head, non-contrast chest, post-contrast chest and post-contrast abdomen-pelvis). Sites contributed data on patients' age, sex and weight, scan factors (tube current and potential), volume CT dose index (CTDIvol) and dose length product (DLP). Data were verified, leading to the exclusion of two sites with missing or incorrect data entries. We estimated overall and site-specific 50th (AD) and 75th (diagnostic reference level [DRL]) percentile CTDIvol and DLP for each CT protocol. Non-normal data were compared using the Kruskal-Wallis test. Sites contributed data from 3,934 children (1,834 females) for different CT exams (head CT 1,568/3,934, 40%; non-contrast chest CT 945/3,934, 24%; post-contrast chest CT 581/3,934, 15%; abdomen-pelvis CT 840/3,934, 21%). There were significant statistical differences in 50th and 75th percentile CTDIvol and DLP values across the participating sites (P<0.001). The 50th and 75th percentile doses for most CT protocols were substantially higher than the corresponding doses reported from the United States of America. Our study demonstrates substantial disparities and variations in pediatric CT examinations performed in multiple sites in Latin America. We will use the collected data to improve scan protocols and perform a follow-up CT study to establish DRLs and ADs based on clinical indications.

Changes in Pulmonary Vascular Resistance and Obstruction Score Following Acute Pulmonary Embolism in Pigs

OBJECTIVES

To investigate the contribution of mechanical obstruction and pulmonary vasoconstriction to pulmonary vascular resistance (PVR) in acute pulmonary embolism (PE) in pigs.

DESIGN

Controlled, animal study.

SETTING

Tertiary university hospital, animal research laboratory.

SUBJECTS

Female Danish slaughter pigs (n = 12, ~60 kg).

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

PE was induced by infusion of autologous blood clots in pigs. CT pulmonary angiograms were performed at baseline, after PE (first experimental day [PEd0]) and the following 2 days (second experimental day [PEd1] and third experimental day [PEd2]), and clot burden quantified by a modified Qanadli Obstruction Score. Hemodynamics were evaluated with left and right heart catheterization and systemic invasive pressures each day before, under, and after treatment with the pulmonary vasodilators sildenafil (0.1 mg/kg) and oxygen (Fio2 40%). PE increased PVR (baseline vs. PEd0: 178 ± 54 vs. 526 ± 160 dynes; p < 0.0001) and obstruction score (baseline vs. PEd0: 0% vs. 45% ± 13%; p < 0.0001). PVR decreased toward baseline at day 1 (baseline vs. PEd1: 178 ± 54 vs. 219 ± 48; p = 0.16) and day 2 (baseline vs. PEd2: 178 ± 54 vs. 201 ± 50; p = 0.51). Obstruction score decreased only slightly at day 1 (PEd0 vs. PEd1: 45% ± 12% vs. 43% ± 14%; p = 0.04) and remained elevated throughout the study (PEd1 vs. PEd2: 43% ± 14% vs. 42% ± 17%; p = 0.74). Sildenafil and oxygen in combination decreased PVR at day 0 (-284 ± 154 dynes; p = 0.0064) but had no effects at day 1 (-8 ± 27 dynes; p = 0.4827) or day 2 (-18 ± 32 dynes; p = 0.0923).

CONCLUSIONS

Pulmonary vasoconstriction, and not mechanical obstruction, was the predominant cause of increased PVR in acute PE in pigs. PVR rapidly declined over the first 2 days after onset despite a persistent mechanical obstruction of the pulmonary circulation from emboli. The findings suggest that treatment with pulmonary vasodilators might only be effective in the acute phase of PE thereby limiting the window for such therapy.

Revisiting the Trustworthiness of Saliency Methods in Radiology AI

Purpose To determine whether saliency maps in radiology artificial intelligence (AI) are vulnerable to subtle perturbations of the input, which could lead to misleading interpretations, using prediction-saliency correlation (PSC) for evaluating the sensitivity and robustness of saliency methods. Materials and Methods In this retrospective study, locally trained deep learning models and a research prototype provided by a commercial vendor were systematically evaluated on 191 229 chest radiographs from the CheXpert dataset and 7022 MR images from a human brain tumor classification dataset. Two radiologists performed a reader study on 270 chest radiograph pairs. A model-agnostic approach for computing the PSC coefficient was used to evaluate the sensitivity and robustness of seven commonly used saliency methods. Results The saliency methods had low sensitivity (maximum PSC, 0.25; 95% CI: 0.12, 0.38) and weak robustness (maximum PSC, 0.12; 95% CI: 0.0, 0.25) on the CheXpert dataset, as demonstrated by leveraging locally trained model parameters. Further evaluation showed that the saliency maps generated from a commercial prototype could be irrelevant to the model output, without knowledge of the model specifics (area under the receiver operating characteristic curve decreased by 8.6% without affecting the saliency map). The human observer studies confirmed that it is difficult for experts to identify the perturbed images; the experts had less than 44.8% correctness. Conclusion Popular saliency methods scored low PSC values on the two datasets of perturbed chest radiographs, indicating weak sensitivity and robustness. The proposed PSC metric provides a valuable quantification tool for validating the trustworthiness of medical AI explainability. Keywords: Saliency Maps, AI Trustworthiness, Dynamic Consistency, Sensitivity, Robustness Supplemental material is available for this article. © RSNA, 2023 See also the commentary by Yanagawa and Sato in this issue.

Low Contrast Volume Protocol in Routine Chest CT Amid the Global Contrast Shortage: A Single Institution Experience

OBJECTIVE

To assess the effectiveness of low contrast volume (LCV) chest CT performed with multiple contrast agents on multivendor CT with varying scanning techniques.

METHODS

The study included 361 patients (65 ± 15 years; M: F 173:188) who underwent LCV chest CT on one of the six 64-256 detector-row CT scanners using single-energy (SECT) or dual-energy (DECT) modes. All patients were scanned with either a fixed-LCV (LCVf, n = 103) or weight-based LCV (LCVw, n = 258) protocol. Two thoracic radiologists independently assessed all LCV CT and patients' prior standard contrast volume (SCV, n = 263) chest CT for optimality of contrast enhancement in thoracic vasculature, cardiac chambers, and in pleuro-parenchymal and mediastinal abnormalities. CT attenuations were recorded in the main pulmonary trunk, ascending, and descending thoracic aorta. To assess the interobserver agreement, pulmonary arterial enhancement was divided into two groups: optimal or suboptimal.

RESULTS

There was no significant difference among patients' BMI (p = 0.883) in the three groups. DECT had a significantly higher aortic arterial enhancement (250 ± 99HU vs 228 ± 76 HU for SECT, p < 0.001). Optimal enhancement was present in 558 of 624 chest CT (89.4%), whereas 66 of 624 chest CT with suboptimal enhancement was noted in 48 of 258 LCVw (18.6%) and 14 of 103 LCVf (13.6%). Most patients with suboptimal enhancement with LCVw injection protocol were overweight/obese (30/48; 62.5%), (p < 0.001).

CONCLUSION

LCV chest CT can be performed across complex multivendor, multicontrast media, multiscanner, and multiprotocol CT practices. However, LCV chest CT examinations can result in suboptimal contrast enhancement in patients with larger body habitus.

Radiologist-Trained AI Model for Identifying Suboptimal Chest-Radiographs

RATIONALE AND OBJECTIVES

Suboptimal chest radiographs (CXR) can limit interpretation of critical findings. Radiologist-trained AI models were evaluated for differentiating suboptimal(sCXR) and optimal(oCXR) chest radiographs.

MATERIALS AND METHODS

Our IRB-approved study included 3278 CXRs from adult patients (mean age 55 ± 20 years) identified from a retrospective search of CXR in radiology reports from 5 sites. A chest radiologist reviewed all CXRs for the cause of suboptimality. The de-identified CXRs were uploaded into an AI server application for training and testing 5 AI models. The training set consisted of 2202 CXRs (n = 807 oCXR; n = 1395 sCXR) while 1076 CXRs (n = 729 sCXR; n = 347 oCXR) were used for testing. Data were analyzed with the Area under the curve (AUC) for the model's ability to classify oCXR and sCXR correctly.

RESULTS

For the two-class classification into sCXR or oCXR from all sites, for CXR with missing anatomy, AI had sensitivity, specificity, accuracy, and AUC of 78%, 95%, 91%, 0.87(95% CI 0.82-0.92), respectively. AI identified obscured thoracic anatomy with 91% sensitivity, 97% specificity, 95% accuracy, and 0.94 AUC (95% CI 0.90-0.97). Inadequate exposure with 90% sensitivity, 93% specificity, 92% accuracy, and AUC of 0.91 (95% CI 0.88-0.95). The presence of low lung volume was identified with 96% sensitivity, 92% specificity, 93% accuracy, and 0.94 AUC (95% CI 0.92-0.96). The sensitivity, specificity, accuracy, and AUC of AI in identifying patient rotation were 92%, 96%, 95%, and 0.94 (95% CI 0.91-0.98), respectively.

CONCLUSION

The radiologist-trained AI models can accurately classify optimal and suboptimal CXRs. Such AI models at the front end of radiographic equipment can enable radiographers to repeat sCXRs when necessary.

Polygenic Risk Score for Cardiovascular Diseases in Artificial Intelligence Paradigm: A Review

Cardiovascular disease (CVD) related mortality and morbidity heavily strain society. The relationship between external risk factors and our genetics have not been well established. It is widely acknowledged that environmental influence and individual behaviours play a significant role in CVD vulnerability, leading to the development of polygenic risk scores (PRS). We employed the PRISMA search method to locate pertinent research and literature to extensively review artificial intelligence (AI)-based PRS models for CVD risk prediction. Furthermore, we analyzed and compared conventional vs. AI-based solutions for PRS. We summarized the recent advances in our understanding of the use of AI-based PRS for risk prediction of CVD. Our study proposes three hypotheses: i) Multiple genetic variations and risk factors can be incorporated into AI-based PRS to improve the accuracy of CVD risk predicting. ii) AI-based PRS for CVD circumvents the drawbacks of conventional PRS calculators by incorporating a larger variety of genetic and non-genetic components, allowing for more precise and individualised risk estimations. iii) Using AI approaches, it is possible to significantly reduce the dimensionality of huge genomic datasets, resulting in more accurate and effective disease risk prediction models. Our study highlighted that the AI-PRS model outperformed traditional PRS calculators in predicting CVD risk. Furthermore, using AI-based methods to calculate PRS may increase the precision of risk predictions for CVD and have significant ramifications for individualized prevention and treatment plans.

DermAI 1.0: A Robust, Generalized, and Novel Attention-Enabled Ensemble-Based Transfer Learning Paradigm for Multiclass Classification of Skin Lesion Images

Skin lesion classification plays a crucial role in dermatology, aiding in the early detection, diagnosis, and management of life-threatening malignant lesions. However, standalone transfer learning (TL) models failed to deliver optimal performance. In this study, we present an attention-enabled ensemble-based deep learning technique, a powerful, novel, and generalized method for extracting features for the classification of skin lesions. This technique holds significant promise in enhancing diagnostic accuracy by using seven pre-trained TL models for classification. Six ensemble-based DL (EBDL) models were created using stacking, softmax voting, and weighted average techniques. Furthermore, we investigated the attention mechanism as an effective paradigm and created seven attention-enabled transfer learning (aeTL) models before branching out to construct three attention-enabled ensemble-based DL (aeEBDL) models to create a reliable, adaptive, and generalized paradigm. The mean accuracy of the TL models is 95.30%, and the use of an ensemble-based paradigm increased it by 4.22%, to 99.52%. The aeTL models' performance was superior to the TL models in accuracy by 3.01%, and aeEBDL models outperformed aeTL models by 1.29%. Statistical tests show significant p-value and Kappa coefficient along with a 99.6% reliability index for the aeEBDL models. The approach is highly effective and generalized for the classification of skin lesions.

Ensemble Deep Learning Derived from Transfer Learning for Classification of COVID-19 Patients on Hybrid Deep-Learning-Based Lung Segmentation: A Data Augmentation and Balancing Framework

BACKGROUND AND MOTIVATION

Lung computed tomography (CT) techniques are high-resolution and are well adopted in the intensive care unit (ICU) for COVID-19 disease control classification. Most artificial intelligence (AI) systems do not undergo generalization and are typically overfitted. Such trained AI systems are not practical for clinical settings and therefore do not give accurate results when executed on unseen data sets. We hypothesize that ensemble deep learning (EDL) is superior to deep transfer learning (TL) in both non-augmented and augmented frameworks.

METHODOLOGY

The system consists of a cascade of quality control, ResNet-UNet-based hybrid deep learning for lung segmentation, and seven models using TL-based classification followed by five types of EDL's. To prove our hypothesis, five different kinds of data combinations (DC) were designed using a combination of two multicenter cohorts-Croatia (80 COVID) and Italy (72 COVID and 30 controls)-leading to 12,000 CT slices. As part of generalization, the system was tested on unseen data and statistically tested for reliability/stability.

RESULTS

Using the K5 (80:20) cross-validation protocol on the balanced and augmented dataset, the five DC datasets improved TL mean accuracy by 3.32%, 6.56%, 12.96%, 47.1%, and 2.78%, respectively. The five EDL systems showed improvements in accuracy of 2.12%, 5.78%, 6.72%, 32.05%, and 2.40%, thus validating our hypothesis. All statistical tests proved positive for reliability and stability.

CONCLUSION

EDL showed superior performance to TL systems for both (a) unbalanced and unaugmented and (b) balanced and augmented datasets for both (i) seen and (ii) unseen paradigms, validating both our hypotheses.

Low concordance between CTPA and echocardiography in identification of right ventricular strain in PERT patients with acute pulmonary embolism

PURPOSE

Right ventricular strain (RVS) is used to risk stratify patients with acute pulmonary embolism (PE) and influence treatment decisions. Guidelines suggest that either computed tomography pulmonary angiography (CTPA) or transthoracic echocardiography (TTE) can be used to assess RVS. We sought to determine how often CTPA and TTE yield discordant results and to assess the test characteristics of CTPA compared to TTE.

METHODS

We analyzed data from a single-center registry of PE cases severe enough to warrant activation of the hospital's Pulmonary Embolism Response Team (PERT). We defined RVS as a right ventricular to left ventricular ratio (RV/LV) ≥ 1 or radiologist's interpretation of RVS on CTPA or as the presence of either RV dilation, hypokinesis, or septal bowing on TTE.

RESULTS

We included 554 patients in our analysis, of whom 333 (60%) had concordant RVS findings on CTPA and TTE. Using TTE as the reference standard, CTPA had a sensitivity of 95% (95% CI 92-97%) and a specificity of 4% (95% CI 2-8%) for identifying RVS.

CONCLUSIONS

In a selected population of patients with acute PE for which PERT was activated, CTPA is highly sensitive but not specific for the detection of RVS when compared to TTE.

Improving Pulmonary Perfusion Assessment by Dynamic Contrast-Enhanced Computed Tomography in an Experimental Lung Injury Model

Introduction: Pulmonary perfusion has been poorly characterized in acute respiratory distress syndrome(ARDS). Optimizing protocols to measure pulmonary blood flow(PBF) via dynamic contrast-enhanced(DCE) computed tomography(CT) could improve understanding of how ARDS alters pulmonary perfusion. In this study, comparative evaluations of injection protocols and tracer-kinetic analysis models were performed based on DCE-CT data measured in ventilated pigs with and without lung injury. Materials and Methods: 10 Yorkshire pigs were anesthetized, intubated, and mechanically ventilated; lung injury was induced by bronchial hydrochloric acid instillation. Each DCE-CT scan was obtained during 30-second end-expiratory breath-hold. Reproducibility of PBF measurements was evaluated in 3 pigs. In 8 pigs, undiluted and diluted Isovue-370 were separately injected to evaluate the effect of contrast viscosity on estimated PBF values. PBF was estimated with the peak-enhancement and the steepest-slope approach. Total-lung PBF was estimated in 2 healthy pigs to compare with cardiac output measured invasively by thermodilution in the pulmonary artery. Results: Repeated measurements in the same animals yielded a good reproducibility of computed PBF maps. Injecting diluted isovue-370 resulted in smaller contrast-time curves in the pulmonary artery(p<0.01) and vein(p<0.01) without substantially diminishing peak signal intensity(p=0.46 in the pulmonary artery) compared to the pure contrast agent since its viscosity is closer to that of blood. As compared to the peak-enhancement model, PBF values estimated by the steepest-slope model with diluted contrast were much closer to the cardiac output(R²=0.82) as compared to the peak-enhancement model. Conclusion: DCE-CT using the steepest-slope model and diluted contrast agent provided reliable quantitative estimates of PBF.

Addressing the Challenges of Implementing Artificial Intelligence Tools in Clinical Practice: Principles From Experience

The multitude of artificial intelligence (AI)-based solutions, vendors, and platforms poses a challenging proposition to an already complex clinical radiology practice. Apart from assessing and ensuring acceptable local performance and workflow fit to improve imaging services, AI tools require multiple stakeholders, including clinical, technical, and financial, who collaborate to move potential deployable applications to full clinical deployment in a structured and efficient manner. Postdeployment monitoring and surveillance of such tools require an infrastructure that ensures proper and safe use. Herein, the authors describe their experience and framework for implementing and supporting the use of AI applications in radiology workflow.

Automatic segmentation and measurement of tracheal collapsibility in tracheomalacia

PURPOSE

To assess feasibility of automated segmentation and measurement of tracheal collapsibility for detecting tracheomalacia on inspiratory and expiratory chest CT images.

METHODS

Our study included 123 patients (age 67 ± 11 years; female: male 69:54) who underwent clinically indicated chest CT examinations in both inspiration and expiration phases. A thoracic radiologist measured anteroposterior length of trachea in inspiration and expiration phase image at the level of maximum collapsibility or aortic arch (in absence of luminal change). Separately, another investigator separately processed the inspiratory and expiratory DICOM CT images with Airway Segmentation component of a commercial COPD software (IntelliSpace Portal, Philips Healthcare). Upon segmentation, the software automatically estimated average lumen diameter (in mm) and lumen area (sq.mm) both along the entire length of trachea and at the level of aortic arch. Data were analyzed with independent t-tests and area under the receiver operating characteristic curve (AUC).

RESULTS

Of the 123 patients, 48 patients had tracheomalacia and 75 patients did not. Ratios of inspiration to expiration phases average lumen area and lumen diameter from the length of trachea had the highest AUC of 0.93 (95% CI = 0.88-0.97) for differentiating presence and absence of tracheomalacia. A decrease of ≥25% in average lumen diameter had sensitivity of 82% and specificity of 87% for detecting tracheomalacia. A decrease of ≥40% in the average lumen area had sensitivity and specificity of 86% for detecting tracheomalacia.

CONCLUSION

Automatic segmentation and measurement of tracheal dimension over the entire tracheal length is more accurate than a single-level measurement for detecting tracheomalacia.

Auto-Detection of Motion Artifacts on CT Pulmonary Angiograms with a Physician-Trained AI Algorithm

Purpose: Motion-impaired CT images can result in limited or suboptimal diagnostic interpretation (with missed or miscalled lesions) and patient recall. We trained and tested an artificial intelligence (AI) model for identifying substantial motion artifacts on CT pulmonary angiography (CTPA) that have a negative impact on diagnostic interpretation. Methods: With IRB approval and HIPAA compliance, we queried our multicenter radiology report database (mPower, Nuance) for CTPA reports between July 2015 and March 2022 for the following terms: "motion artifacts", "respiratory motion", "technically inadequate", and "suboptimal" or "limited exam". All CTPA reports were from two quaternary (Site A, n = 335; B, n = 259) and a community (C, n = 199) healthcare sites. A thoracic radiologist reviewed CT images of all positive hits for motion artifacts (present or absent) and their severity (no diagnostic effect or major diagnostic impairment). Coronal multiplanar images from 793 CTPA exams were de-identified and exported offline into an AI model building prototype (Cognex Vision Pro, Cognex Corporation) to train an AI model to perform two-class classification ("motion" or "no motion") with data from the three sites (70% training dataset, n = 554; 30% validation dataset, n = 239). Separately, data from Site A and Site C were used for training and validating; testing was performed on the Site B CTPA exams. A five-fold repeated cross-validation was performed to evaluate the model performance with accuracy and receiver operating characteristics analysis (ROC). Results: Among the CTPA images from 793 patients (mean age 63 ± 17 years; 391 males, 402 females), 372 had no motion artifacts, and 421 had substantial motion artifacts. The statistics for the average performance of the AI model after five-fold repeated cross-validation for the two-class classification included 94% sensitivity, 91% specificity, 93% accuracy, and 0.93 area under the ROC curve (AUC: 95% CI 0.89-0.97). Conclusion: The AI model used in this study can successfully identify CTPA exams with diagnostic interpretation limiting motion artifacts in multicenter training and test datasets. Clinical relevance: The AI model used in the study can help alert technologists about the presence of substantial motion artifacts on CTPA, where a repeat image acquisition can help salvage diagnostic information.

Performance of threshold-based stone segmentation and radiomics for determining the composition of kidney stones from single-energy CT

PURPOSE

Knowledge of kidney stone composition can help in patient management; urine composition analysis and dual-energy CT are frequently used to assess stone type. We assessed if threshold-based stone segmentation and radiomics can determine the composition of kidney stones from single-energy, non-contrast abdomen-pelvis CT.

METHODS

With IRB approval, we identified 218 consecutive patients (mean age 64 ± 13 years; male:female 138:80) with the presence of kidney stones on non-contrast, abdomen-pelvis CT and surgical or biochemical proof of their stone composition. CT examinations were performed on one of the seven multidetector-row scanners from four vendors (GE, Philips, Siemens, Toshiba). Deidentified CT images were processed with a radiomics prototype (Frontier, Siemens Healthineers) to segment the entire kidney volumes with an AI-based organ segmentation tool. We applied a threshold of 130 HU to isolate stones in the segmented kidneys and to estimate radiomics over the segmented stone volume. A coinvestigator verified kidney stone segmentation and adjusted the volume of interest to include the entire stone volume when necessary. We applied multiple logistic regression tests with precision recall plots to obtain area under the curve (AUC) using a built-in R statistical program.

RESULTS

The threshold-based stone segmentation successfully isolated kidney stones (uric acid: n = 102 patients, calcium oxalate/phosphate: n = 116 patients) in all patients. Radiomics differentiated between calcium and uric acid stones with an AUC of 0.78 (p < 0.01, 95% CI 0.73-0.83), 0.79 sensitivity, and 0.90 specificity regardless of CT vendors (GE CT: AUC = 0.82, p < 0.01, 95% CI 0.740-0896; Siemens CT: AUC = 0.77, 95% CI 0.700-0.846, p < 0.01).

CONCLUSION

Automated threshold-based stone segmentation and radiomics can differentiate between calcium oxalate/phosphate and urate stones from non-contrast, single-energy abdomen CT.

Evaluation of an Artificial Intelligence Model for Detection of Pneumothorax and Tension Pneumothorax in Chest Radiographs

IMPORTANCE

Early detection of pneumothorax, most often via chest radiography, can help determine need for emergent clinical intervention. The ability to accurately detect and rapidly triage pneumothorax with an artificial intelligence (AI) model could assist with earlier identification and improve care.

OBJECTIVE

To compare the accuracy of an AI model vs consensus thoracic radiologist interpretations in detecting any pneumothorax (incorporating both nontension and tension pneumothorax) and tension pneumothorax.

DESIGN, SETTING, AND PARTICIPANTS

This diagnostic study was a retrospective standalone performance assessment using a data set of 1000 chest radiographs captured between June 1, 2015, and May 31, 2021. The radiographs were obtained from patients aged at least 18 years at 4 hospitals in the Mass General Brigham hospital network in the United States. Included radiographs were selected using 2 strategies from all chest radiography performed at the hospitals, including inpatient and outpatient. The first strategy identified consecutive radiographs with pneumothorax through a manual review of radiology reports, and the second strategy identified consecutive radiographs with tension pneumothorax using natural language processing. For both strategies, negative radiographs were selected by taking the next negative radiograph acquired from the same radiography machine as each positive radiograph. The final data set was an amalgamation of these processes. Each radiograph was interpreted independently by up to 3 radiologists to establish consensus ground-truth interpretations. Each radiograph was then interpreted by the AI model for the presence of pneumothorax and tension pneumothorax. This study was conducted between July and October 2021, with the primary analysis performed between October and November 2021.

MAIN OUTCOMES AND MEASURES

The primary end points were the areas under the receiver operating characteristic curves (AUCs) for the detection of pneumothorax and tension pneumothorax. The secondary end points were the sensitivities and specificities for the detection of pneumothorax and tension pneumothorax.

RESULTS

The final analysis included radiographs from 985 patients (mean [SD] age, 60.8 [19.0] years; 436 [44.3%] female patients), including 307 patients with nontension pneumothorax, 128 patients with tension pneumothorax, and 550 patients without pneumothorax. The AI model detected any pneumothorax with an AUC of 0.979 (95% CI, 0.970-0.987), sensitivity of 94.3% (95% CI, 92.0%-96.3%), and specificity of 92.0% (95% CI, 89.6%-94.2%) and tension pneumothorax with an AUC of 0.987 (95% CI, 0.980-0.992), sensitivity of 94.5% (95% CI, 90.6%-97.7%), and specificity of 95.3% (95% CI, 93.9%-96.6%).

CONCLUSIONS AND RELEVANCE

These findings suggest that the assessed AI model accurately detected pneumothorax and tension pneumothorax in this chest radiograph data set. The model's use in the clinical workflow could lead to earlier identification and improved care for patients with pneumothorax.

Complex Relationship Between Artificial Intelligence and CT Radiation Dose

Concerns over need for CT radiation dose optimization and reduction led to improved scanner efficiency and introduction of several reconstruction techniques and image processing-based software. The latest technologies use artificial intelligence (AI) for CT dose optimization and image quality improvement. While CT dose optimization has and can benefit from AI, variations in scanner technologies, reconstruction methods, and scan protocols can lead to substantial variations in radiation doses and image quality across and within different scanners. These variations in turn can influence performance of AI algorithms being deployed for tasks such as detection, segmentation, characterization, and quantification. We review the complex relationship between AI and CT radiation dose.

Multicenter Study on COVID-19 Lung Computed Tomography Segmentation with varying Glass Ground Opacities using Unseen Deep Learning Artificial Intelligence Paradigms: COVLIAS 1.0 Validation

Variations in COVID-19 lesions such as glass ground opacities (GGO), consolidations, and crazy paving can compromise the ability of solo-deep learning (SDL) or hybrid-deep learning (HDL) artificial intelligence (AI) models in predicting automated COVID-19 lung segmentation in Computed Tomography (CT) from unseen data leading to poor clinical manifestations. As the first study of its kind, “COVLIAS 1.0-Unseen” proves two hypotheses, (i) contrast adjustment is vital for AI, and (ii) HDL is superior to SDL. In a multicenter study, 10,000 CT slices were collected from 72 Italian (ITA) patients with low-GGO, and 80 Croatian (CRO) patients with high-GGO. Hounsfield Units (HU) were automatically adjusted to train the AI models and predict from test data, leading to four combinations—two Unseen sets: (i) train-CRO:test-ITA, (ii) train-ITA:test-CRO, and two Seen sets: (iii) train-CRO:test-CRO, (iv) train-ITA:test-ITA. COVILAS used three SDL models: PSPNet, SegNet, UNet and six HDL models: VGG-PSPNet, VGG-SegNet, VGG-UNet, ResNet-PSPNet, ResNet-SegNet, and ResNet-UNet. Two trained, blinded senior radiologists conducted ground truth annotations. Five types of performance metrics were used to validate COVLIAS 1.0-Unseen which was further benchmarked against MedSeg, an open-source web-based system. After HU adjustment for DS and JI, HDL (Unseen AI) > SDL (Unseen AI) by 4% and 5%, respectively. For CC, HDL (Unseen AI) > SDL (Unseen AI) by 6%. The COVLIAS-MedSeg difference was < 5%, meeting regulatory guidelines.Unseen AI was successfully demonstrated using automated HU adjustment. HDL was found to be superior to SDL.

Effect of technologist and patient attributes on centering for body CT examinations: Influence of cultural and ethnic factors

There are no published data on the effect of patient and technologist gender and ethnicity attributes on off-centering in CT. Therefore, we assessed the impact of patient and technologist variations on off-centering patients undergoing body CT. With institutional review board approval, our retrospective study included 1000 consecutive adult patients (age ranged 22–96 years; 756 males: 244 females) who underwent chest or abdomen CT examinations. We recorded patient (age, gender, nationality, body weight, height,), technologist gender, and scan-related (scanner vendor, body region imaged, scan length, CT dose index volume, dose length product) information. Lateral and anteroposterior (AP) diameters were recorded to calculate effective diameter and size-specific dose estimate (SSDE). Off-centering represented the distance between the anterior-posterior centers of the scan field of view and the patient at the level of carina (for chest CT) and iliac crest (for abdomen CT). About 76% of the patients (760/1000) were off-centered with greater off-centering for chest (22 mm) than for abdomen (15 mm). Although ethnicity or patient gender was not a significant determinant of off-centering, technologist-patient gender mismatch was associated with a significantly greater frequency of off-centering (p<0.001). Off-centering below the gantry isocenter was twice as common as off-centering above the gantry isocenter (p<0.001). The latter occurred more frequently in larger patients and was associated with higher radiation doses than those centered below the isocenter (p<0.001). Technologists’ years of experience and patient factors profoundly affect the presence and extent of off-centering for both chest and abdomen CTs. Larger patients are more often off-centered than smaller patients.

Evaluating cardiopulmonary function following acute pulmonary embolism

INTRODUCTION

Pulmonary embolism is a common cause of cardiopulmonary mortality and morbidity worldwide. Survivors of acute pulmonary embolism may experience dyspnea, report reduced exercise capacity, or develop overt pulmonary hypertension. Clinicians must be alert for these phenomena and appreciate the modalities and investigations available for evaluation.

AREAS COVERED

In this review, the current understanding of available contemporary imaging and physiologic modalities is discussed, based on available literature and professional society guidelines. The purpose of the review is to provide clinicians with an overview of these modalities, their strengths and disadvantages, and how and when these investigations can support the clinical work-up of patients post-pulmonary embolism.

EXPERT OPINION

Echocardiography is a first test in symptomatic patients post-pulmonary embolism, with ventilation/perfusion scanning vital to determination of whether there is chronic residual emboli. The role of computed tomography and magnetic resonance in assessing the pulmonary arterial tree in post-pulmonary embolism patients is evolving. Functional testing, in particular cardiopulmonary exercise testing, is emerging as an important modality to quantify and determine cause of functional limitation. It is possible that future investigations of the post-pulmonary embolism recovery period will better inform treatment decisions for acute pulmonary embolism patients.

Association of Artificial Intelligence-Aided Chest Radiograph Interpretation With Reader Performance and Efficiency

IMPORTANCE

The efficient and accurate interpretation of radiologic images is paramount.

OBJECTIVE

To evaluate whether a deep learning-based artificial intelligence (AI) engine used concurrently can improve reader performance and efficiency in interpreting chest radiograph abnormalities.

DESIGN, SETTING, AND PARTICIPANTS

This multicenter cohort study was conducted from April to November 2021 and involved radiologists, including attending radiologists, thoracic radiology fellows, and residents, who independently participated in 2 observer performance test sessions. The sessions included a reading session with AI and a session without AI, in a randomized crossover manner with a 4-week washout period in between. The AI produced a heat map and the image-level probability of the presence of the referrable lesion. The data used were collected at 2 quaternary academic hospitals in Boston, Massachusetts: Beth Israel Deaconess Medical Center (The Medical Information Mart for Intensive Care Chest X-Ray [MIMIC-CXR]) and Massachusetts General Hospital (MGH).

MAIN OUTCOMES AND MEASURES

The ground truths for the labels were created via consensual reading by 2 thoracic radiologists. Each reader documented their findings in a customized report template, in which the 4 target chest radiograph findings and the reader confidence of the presence of each finding was recorded. The time taken for reporting each chest radiograph was also recorded. Sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC) were calculated for each target finding.

RESULTS

A total of 6 radiologists (2 attending radiologists, 2 thoracic radiology fellows, and 2 residents) participated in the study. The study involved a total of 497 frontal chest radiographs-247 from the MIMIC-CXR data set (demographic data for patients were not available) and 250 chest radiographs from MGH (mean [SD] age, 63 [16] years; 133 men [53.2%])-from adult patients with and without 4 target findings (pneumonia, nodule, pneumothorax, and pleural effusion). The target findings were found in 351 of 497 chest radiographs. The AI was associated with higher sensitivity for all findings compared with the readers (nodule, 0.816 [95% CI, 0.732-0.882] vs 0.567 [95% CI, 0.524-0.611]; pneumonia, 0.887 [95% CI, 0.834-0.928] vs 0.673 [95% CI, 0.632-0.714]; pleural effusion, 0.872 [95% CI, 0.808-0.921] vs 0.889 [95% CI, 0.862-0.917]; pneumothorax, 0.988 [95% CI, 0.932-1.000] vs 0.792 [95% CI, 0.756-0.827]). AI-aided interpretation was associated with significantly improved reader sensitivities for all target findings, without negative impacts on the specificity. Overall, the AUROCs of readers improved for all 4 target findings, with significant improvements in detection of pneumothorax and nodule. The reporting time with AI was 10% lower than without AI (40.8 vs 36.9 seconds; difference, 3.9 seconds; 95% CI, 2.9-5.2 seconds; P < .001).

CONCLUSIONS AND RELEVANCE

These findings suggest that AI-aided interpretation was associated with improved reader performance and efficiency for identifying major thoracic findings on a chest radiograph.

Characterization of Benign and Malignant Pancreatic Lesions with DECT Quantitative Metrics and Radiomics

RATIONALE AND OBJECTIVES

To compare dual energy CT (DECT) quantitative metrics and radiomics for differentiating benign and malignant pancreatic lesions on contrast enhanced abdomen CT.

MATERIALS AND METHODS

Our study included 103 patients who underwent contrast-enhanced DECT for assessing focal pancreatic lesions at one of the two hospitals (Site A: age 68 ± 12 yrs; malignant = 41, benign = 18; Site B: age 46 ± 2 yrs; malignant = 23, benign = 21). All malignant lesions had histologic confirmation, and benign lesions were stable on follow up CT (>12 months) or had characteristic benign features on MRI. Arterial-phase, low- and high-kV DICOM images were processed with the DECT Tumor Analysis (DETA) to obtain DECT quantitative metrics such as HU, iodine and water content from a region of interest (ROI) over focal pancreatic lesions. Separately, we obtained DECT radiomics from the same ROI. Data were analyzed with multiple logistic regression and receiver operating characteristics to generate area under the curve (AUC) for best predictive variables.

RESULTS

DECT quantitative metrics and radiomics had AUCs of 0.98-0.99 at site A and 0.89-0.94 at site B data for classifying benign and malignant pancreatic lesions. There was no significant difference in the AUCs and accuracies of DECT quantitative metrics and radiomics from lesion rims and volumes among patients at both sites (p > 0.05). Supervised learning-based model with data from the two sites demonstrated best AUCs of 0.94 (DECT radiomics) and 0.90 (DECT quantitative metrics) for characterizing pancreatic lesions as benign or malignant.

CONCLUSION

Compared to complex DECT radiomics, quantitative DECT information provide a simpler but accurate method of differentiating benign and malignant pancreatic lesions.

Spectral segmentation and radiomic features predict carotid stenosis and ipsilateral ischemic burden from DECT angiography

PURPOSE The purpose of this study is to compare spectral segmentation, spectral radiomic, and single- energy radiomic features in the assessment of internal and common carotid artery (ICA/CCA) stenosis and prediction of surgical outcome. METHODS Our ethical committee-approved, Health Insurance Portability and Accountability Act (HIPAA)- compliant study included 85 patients (mean age, 73 ± 10 years; male : female, 56 : 29) who under- went contrast-enhanced, dual-source dual-energy CT angiography (DECTA) (Siemens Definition Flash) of the neck for assessing ICA/CCA stenosis. Patients with a prior surgical or interventional treatment of carotid stenosis were excluded. Two radiologists graded the severity of carotid ste- nosis on DECTA images as mild (<50% luminal narrowing), moderate (50%-69%), and severe (>70%) stenosis. Thin-section, low- and high-kV DICOM images from the arterial phase acquisi- tion were processed with a dual-energy CT prototype (DTA, eXamine, Siemens Healthineers) to generate spectral segmentation and radiomic features over regions of interest along the entire length (volume) and separately at a single-section with maximum stenosis. Multiple logistic regressions and area under the receiver operating characteristic curve (AUC) were used for data analysis. RESULTS Among 85 patients, 22 ICA/CCAs had normal luminal dimensions and 148 ICA/CCAs had luminal stenosis (mild stenosis: 51, moderate: 38, severe: 59). For differentiating non-severe and severe ICA/CCA stenosis, radiomic features (volume: AUC=0.94, 95% CI 0.88-0.96; section: AUC=0.92, 95% CI 0.86-0.93) were significantly better than spectral segmentation features (volume: AUC = 0.86, 95% CI 0.74-0.87; section: AUC = 0.68, 95% CI 0.66-0.78) (P < .001). Spectral radiomic features predicted revascularization procedure (AUC = 0.77) and the presence of ipsilateral intra- cranial ischemic changes (AUC = 0.76). CONCLUSION Spectral segmentation and radiomic features from DECTA can differentiate patients with differ- ent luminal ICA/CCA stenosis grades.

FDA-regulated AI Algorithms: Trends, Strengths, and Gaps of Validation Studies

RATIONALE AND OBJECTIVES

To assess key trends, strengths, and gaps in validation studies of the Food and Drug Administration (FDA)-regulated imaging-based artificial intelligence/machine learning (AI/ML) algorithms.

MATERIALS AND METHODS

We audited publicly available details of regulated AI/ML algorithms in imaging from 2008 until April 2021. We reviewed 127 regulated software (118 AI/ML) to classify information related to their parent company, subspecialty, body area and specific anatomy type, imaging modality, date of FDA clearance, indications for use, target pathology (such as trauma) and findings (such as fracture), technique (CAD triage, CAD detection and/or characterization, CAD acquisition or improvement, and image processing/quantification), product performance, presence, type, strength and availability of clinical validation data. Pertaining to validation data, where available, we recorded the number of patients or studies included, sensitivity, specificity, accuracy, and/or receiver operating characteristic area under the curve, along with information on ground-truthing of use-cases. Data were analyzed with pivot tables and charts for descriptive statistics and trends.

RESULTS

We noted an increasing number of FDA-regulated AI/ML from 2008 to 2021. Seventeen (17/118) regulated AI/ML algorithms posted no validation claims or data. Just 9/118 reviewed AI/ML algorithms had a validation dataset sizes of over 1000 patients. The most common type of AI/ML included image processing/quantification (IPQ; n = 59/118), and triage (CADt; n = 27/118). Brain, breast, and lungs dominated the targeted body regions of interest.

CONCLUSION

Insufficient public information on validation datasets in several FDA-regulated AI/ML algorithms makes it difficult to justify clinical applications since their generalizability and presence of bias cannot be inferred.

National reference levels of CT procedures dedicated for treatment planning in radiation oncology

OBJECTIVE

To present results of the first national survey on reference levels of CT imaging performed for the treatment planning purposes in radiation oncology in Croatia.

METHODS

Data for CT protocols of five anatomical regions including head, head and neck, pelvis, breast, and thorax were collected at eight radiation oncology departments in Croatia. Data included volume CT dose index (CTDI_vol), dose-length product (DLP), scan length and set of acquisition and reconstruction parameters. Data on a total of 600 patients were collected. Median values of scan length, DLP and CTDI_vol were calculated for each acquisition protocol. Third quartiles of the median CTDI_vol and DLP values were proposed as the national radiotherapy planning reference levels (RPRL).

RESULTS

The largest CoV were assessed for RT Breast (63.8% for CTDI_vol), RT Thorax (79.7% for DLP) and RT H&N (21.2% for scan length). RT Head had the lowest CoV for CTDI_vol (1,9%) and DLP (17,2%), while RT Breast had the lowest coefficient of variation for scan length (12.8%). Proposed national RPRLs are: for RT Head CTDI_vol16cm = 62 mGy and DLP_16cm = 1738 mGy.cm; for RT H&N CTDI_vol16cm = 35 mGy and DLP_16cm = 1444 mGy.cm; for RT Breast CTDI_vol32cm = 16 mGy and DLP_32cm = 731 mGy.cm; for RT Thorax CTDI_vol32cm = 17 mGy and DLP_32cm = 865 mGy.cm; for RT Pelvis CTDI_vol32cm = 20 mGy and DLP_32cm = 1133 mGy.cm.

CONCLUSIONS

Results of this study show variations in CT imaging for treatment planning practice at the national level which call for optimization of procedures.

Four Types of Multiclass Frameworks for Pneumonia Classification and Its Validation in X-ray Scans Using Seven Types of Deep Learning Artificial Intelligence Models

Background and Motivation: The novel coronavirus causing COVID-19 is exceptionally contagious, highly mutative, decimating human health and life, as well as the global economy, by consistent evolution of new pernicious variants and outbreaks. The reverse transcriptase polymerase chain reaction currently used for diagnosis has major limitations. Furthermore, the multiclass lung classification X-ray systems having viral, bacterial, and tubercular classes—including COVID-19—are not reliable. Thus, there is a need for a robust, fast, cost-effective, and easily available diagnostic method. Method: Artificial intelligence (AI) has been shown to revolutionize all walks of life, particularly medical imaging. This study proposes a deep learning AI-based automatic multiclass detection and classification of pneumonia from chest X-ray images that are readily available and highly cost-effective. The study has designed and applied seven highly efficient pre-trained convolutional neural networks—namely, VGG16, VGG19, DenseNet201, Xception, InceptionV3, NasnetMobile, and ResNet152—for classification of up to five classes of pneumonia. Results: The database consisted of 18,603 scans with two, three, and five classes. The best results were using DenseNet201, VGG16, and VGG16, respectively having accuracies of 99.84%, 96.7%, 92.67%; sensitivity of 99.84%, 96.63%, 92.70%; specificity of 99.84, 96.63%, 92.41%; and AUC of 1.0, 0.97, 0.92 (p < 0.0001 for all), respectively. Our system outperformed existing methods by 1.2% for the five-class model. The online system takes <1 s while demonstrating reliability and stability. Conclusions: Deep learning AI is a powerful paradigm for multiclass pneumonia classification.

CHEST CT USAGE IN COVID-19 PNEUMONIA: MULTICENTER STUDY ON RADIATION DOSES AND DIAGNOSTIC QUALITY IN BRAZIL

We assessed variations in chest CT usage, radiation dose and image quality in COVID-19 pneumonia. Our study included all chest CT exams performed in 533 patients from 6 healthcare sites from Brazil. We recorded patients' age, gender and body weight and the information number of CT exams per patient, scan parameters and radiation doses (volume CT dose index-CTDIvol and dose length product-DLP). Six radiologists assessed all chest CT exams for the type of pulmonary findings and classified CT appearance of COVID-19 pneumonia as typical, indeterminate, atypical or negative. In addition, each CT was assessed for diagnostic quality (optimal or suboptimal) and presence of artefacts. Artefacts were frequent (367/841), often related to respiratory motion (344/367 chest CT exams with artefacts) and resulted in suboptimal evaluation in mid-to-lower lungs (176/344) or the entire lung (31/344). There were substantial differences in CT usage, patient weight, CTDIvol and DLP across the participating sites.

COVLIAS 1.0 vs. MedSeg: Artificial Intelligence-Based Comparative Study for Automated COVID-19 Computed Tomography Lung Segmentation in Italian and Croatian Cohorts

(1) Background: COVID-19 computed tomography (CT) lung segmentation is critical for COVID lung severity diagnosis. Earlier proposed approaches during 2020–2021 were semiautomated or automated but not accurate, user-friendly, and industry-standard benchmarked. The proposed study compared the COVID Lung Image Analysis System, COVLIAS 1.0 (GBTI, Inc., and AtheroPoint^TM, Roseville, CA, USA, referred to as COVLIAS), against MedSeg, a web-based Artificial Intelligence (AI) segmentation tool, where COVLIAS uses hybrid deep learning (HDL) models for CT lung segmentation. (2) Materials and Methods: The proposed study used 5000 ITALIAN COVID-19 positive CT lung images collected from 72 patients (experimental data) that confirmed the reverse transcription-polymerase chain reaction (RT-PCR) test. Two hybrid AI models from the COVLIAS system, namely, VGG-SegNet (HDL 1) and ResNet-SegNet (HDL 2), were used to segment the CT lungs. As part of the results, we compared both COVLIAS and MedSeg against two manual delineations (MD 1 and MD 2) using (i) Bland–Altman plots, (ii) Correlation coefficient (CC) plots, (iii) Receiver operating characteristic curve, and (iv) Figure of Merit and (v) visual overlays. A cohort of 500 CROATIA COVID-19 positive CT lung images (validation data) was used. A previously trained COVLIAS model was directly applied to the validation data (as part of Unseen-AI) to segment the CT lungs and compare them against MedSeg. (3) Result: For the experimental data, the four CCs between COVLIAS (HDL 1) vs. MD 1, COVLIAS (HDL 1) vs. MD 2, COVLIAS (HDL 2) vs. MD 1, and COVLIAS (HDL 2) vs. MD 2 were 0.96, 0.96, 0.96, and 0.96, respectively. The mean value of the COVLIAS system for the above four readings was 0.96. CC between MedSeg vs. MD 1 and MedSeg vs. MD 2 was 0.98 and 0.98, respectively. Both had a mean value of 0.98. On the validation data, the CC between COVLIAS (HDL 1) vs. MedSeg and COVLIAS (HDL 2) vs. MedSeg was 0.98 and 0.99, respectively. For the experimental data, the difference between the mean values for COVLIAS and MedSeg showed a difference of <2.5%, meeting the standard of equivalence. The average running times for COVLIAS and MedSeg on a single lung CT slice were ~4 s and ~10 s, respectively. (4) Conclusions: The performances of COVLIAS and MedSeg were similar. However, COVLIAS showed improved computing time over MedSeg.

Inter-Variability Study of COVLIAS 1.0: Hybrid Deep Learning Models for COVID-19 Lung Segmentation in Computed Tomography

Background: For COVID-19 lung severity, segmentation of lungs on computed tomography (CT) is the first crucial step. Current deep learning (DL)-based Artificial Intelligence (AI) models have a bias in the training stage of segmentation because only one set of ground truth (GT) annotations are evaluated. We propose a robust and stable inter-variability analysis of CT lung segmentation in COVID-19 to avoid the effect of bias. Methodology: The proposed inter-variability study consists of two GT tracers for lung segmentation on chest CT. Three AI models, PSP Net, VGG-SegNet, and ResNet-SegNet, were trained using GT annotations. We hypothesized that if AI models are trained on the GT tracings from multiple experience levels, and if the AI performance on the test data between these AI models is within the 5% range, one can consider such an AI model robust and unbiased. The K5 protocol (training to testing: 80%:20%) was adapted. Ten kinds of metrics were used for performance evaluation. Results: The database consisted of 5000 CT chest images from 72 COVID-19-infected patients. By computing the coefficient of correlations (CC) between the output of the two AI models trained corresponding to the two GT tracers, computing their differences in their CC, and repeating the process for all three AI-models, we show the differences as 0%, 0.51%, and 2.04% (all < 5%), thereby validating the hypothesis. The performance was comparable; however, it had the following order: ResNet-SegNet > PSP Net > VGG-SegNet. Conclusions: The AI models were clinically robust and stable during the inter-variability analysis on the CT lung segmentation on COVID-19 patients.

Deep Interactive Denoiser (DID) for X-Ray Computed Tomography

Low-dose computed tomography (LDCT) is desirable for both diagnostic imaging and image-guided interventions. Denoisers are widely used to improve the quality of LDCT. Deep learning (DL)-based denoisers have shown state-of-the-art performance and are becoming mainstream methods. However, there are two challenges to using DL-based denoisers: 1) a trained model typically does not generate different image candidates with different noise-resolution tradeoffs, which are sometimes needed for different clinical tasks; and 2) the model's generalizability might be an issue when the noise level in the testing images differs from that in the training dataset. To address these two challenges, in this work, we introduce a lightweight optimization process that can run on top of any existing DL-based denoiser during the testing phase to generate multiple image candidates with different noise-resolution tradeoffs suitable for different clinical tasks in real time. Consequently, our method allows users to interact with the denoiser to efficiently review various image candidates and quickly pick the desired one; thus, we termed this method deep interactive denoiser (DID). Experimental results demonstrated that DID can deliver multiple image candidates with different noise-resolution tradeoffs and shows great generalizability across various network architectures, as well as training and testing datasets with various noise levels.

Artificial Intelligence has Similar Performance to Subjective Assessment of Emphysema Severity on Chest CT

RATIONALE AND OBJECTIVES

To compare an artificial intelligence (AI)-based prototype and subjective grading for predicting disease severity in patients with emphysema.

METHODS

Our IRB approved HIPAA-compliant study included 113 adults (71±8 years; 47 females, 66 males) who had both non-contrast chest CT and pulmonary function tests performed within a span of 2 months. The disease severity was classified based on the forced expiratory volume in 1 second (FEV1 as % of predicted) into mild, moderate, and severe. 2 thoracic radiologists (RA), blinded to the clinical and AI results, graded severity of emphysema on a 5-point scale suggested by the Fleischner Society for each lobe. The whole lung scores were derived from the summation of lobar scores. Thin-section CT images were processed with the AI-Rad Companion Chest prototype (Siemens Healthineers) to quantify low attenuation areas (LAA < - 950 HU) in whole lung and each lobe separately. Bronchial abnormality was assessed by both radiologists and a fully automated software (Philips Healthcare).

RESULTS

Both AI (AUC of 0.77; 95% CI: 0.68 - 0.85) and RA (AUC: 0.76, 95% CI: 0.65 - 0.84) emphysema quantification could differentiate mild, moderate, and severe disease based on FEV1. There was a strong positive correlation between AI and RA (r = 0.72 - 0.80; p <0.001). The combination of emphysema and bronchial abnormality quantification from radiologists' and AI assessment could differentiate between different severities with AUC of 0.80 - 0.82 and 0.87, respectively.

CONCLUSION

The assessed AI-prototypes can predict the disease severity in patients with emphysema with the same predictive value as the radiologists.

PRACTICAL CHALLENGES WITH IMAGING COVID-19 IN BRAZIL: MITIGATION IN AND BEYOND THE PANDEMIC

Computed tomography (CT) provides useful information in patients with known or suspected COVID-19 infection. However, there are substantial variations and challenges in scanner technologies and scan practices that have negative effect on the image quality and can increase radiation dose associated with CT.

OBJECTIVE

In this article, we present major issues and challenges with use of CT at five Brazilian CT facilities for imaging patients with known or suspected COVID-19 infection and offer specific mitigating strategies.

METHODS

Observational, retrospective and prospective study of five CT facilities from different states and regions of Brazil, with approval of research and ethics committees.

RESULTS

The most important issues include frequent use of CT, lack of up-to-date and efficient scanner technologies, over-scanning and patient off-centring. Mitigating strategies can include updating scanner technology and improving scan practices.

Comparison of Baseline, Bone-Subtracted, and Enhanced Chest Radiographs for Detection of Pneumothorax

Purpose:

To assess and compare detectability of pneumothorax on unprocessed baseline, single-energy, bone-subtracted, and enhanced frontal chest radiographs (chest X-ray, CXR).

Method and Materials:

Our retrospective institutional review board–approved study included 202 patients (mean age 53 ± 24 years; 132 men, 70 women) who underwent frontal CXR and had trace, moderate, large, or tension pneumothorax. All patients (except those with tension pneumothorax) had concurrent chest computed tomography (CT). Two radiologists reviewed the CXR and chest CT for pneumothorax on baseline CXR (ground truth). All baseline CXR were processed to generate bone-subtracted and enhanced images (ClearRead X-ray). Four radiologists (R1-R4) assessed the baseline, bone-subtracted, and enhanced images and recorded the presence of pneumothorax (side, size, and confidence for detection) for each image type. Area under the curve (AUC) was calculated with receiver operating characteristic analyses to determine the accuracy of pneumothorax detection.

Results:

Bone-subtracted images (AUC: 0.89-0.97) had the lowest accuracy for detection of pneumothorax compared to the baseline (AUC: 0.94-0.97) and enhanced (AUC: 0.96-0.99) radiographs ( P < .01). Most false-positive and false-negative pneumothoraces were detected on the bone-subtracted images and the least numbers on the enhanced radiographs. Highest detection rates and confidence were noted for the enhanced images (empiric AUC for R1-R4 0.96-0.99).

Conclusion:

Enhanced CXRs are superior to bone-subtracted and unprocessed radiographs for detection of pneumothorax.

Clinical Relevance/Application:

Enhanced CXRs improve detection of pneumothorax over unprocessed images; bone-subtracted images must be cautiously reviewed to avoid false negatives.

Iterative material decomposition for spectral CT using self-supervised Noise2Noise prior

Compared to conventional computed tomography (CT), spectral CT can provide the capability of material decomposition, which can be used in many clinical diagnosis applications. However, the decomposed images can be very noisy due to the dose limit in CT scanning and the noise magnification of the material decomposition process. To alleviate this situation, we proposed an iterative one-step inversion material decomposition algorithm with a Noise2Noise prior. The algorithm estimated material images directly from projection data and used a Noise2Noise prior for denoising. In contrast to supervised deep learning methods, the designed Noise2Noise prior was built based on self-supervised learning and did not need external data for training. In our method, the data consistency term and the Noise2Noise network were alternatively optimized in the iterative framework, respectively, using a separable quadratic surrogate (SQS) and the Adam algorithm. The proposed iterative algorithm was validated and compared to other methods on simulated spectral CT data, preclinical photon-counting CT data and clinical dual-energy CT data. Quantitative analysis showed that our proposed method performs promisingly on noise suppression and structure detail recovery.

Prediction of Coronary Calcification and Stenosis: Role of Radiomics From Low-Dose CT

RATIONALE AND OBJECTIVES

We aimed to assess relationship between single-click, whole heart radiomics from low-dose computed tomography (LDCT) for lung cancer screening with coronary artery calcification and stenosis.

MATERIALS AND METHODS

The institutional review board-approved, retrospective study included all 106 patients (68 men, 38 women, mean age 64 ± 7 years) who underwent both LDCT for lung cancer screening and had calcium scoring and coronary computed tomography angiography in our institution. We recorded the clinical variables including patients' demographics, smoking history, family history, and lipid profiles. Coronary calcium scores and grading of coronary stenosis were recorded from the radiology information system. We calculated the multiethnic scores for atherosclerosis risk scores to obtain 10-year coronary heart disease (MESA 10-Y CHD) risk of cardiovascular disease for all patients. Deidentified LDCT exams were exported to a Radiomics prototype for automatic heart segmentation, and derivation of radiomics. Data were analyzed using multiple logistic regression and kernel Fisher discriminant analyses.

RESULTS

Whole heart radiomics were better than the clinical variables for differentiating subjects with different Agatston scores (≤400 and >400) (area under the curve [AUC] 0.92 vs 0.69). Prediction of coronary stenosis and MESA 10-Y CHD risk was better on whole heart radiomics (AUC:0.86-0.87) than with clinical variables (AUC:0.69-0.79). Addition of clinical variables or visual assessment of coronary calcification from LDCT to whole heart radiomics resulted in a modest change in the AUC.

CONCLUSION

Single-click, whole heart radiomics obtained from LDCT for lung cancer screening can differentiate patients with different Agatston and MESA risk scores for cardiovascular diseases.

Comparison of deep learning, radiomics and subjective assessment of chest CT findings in SARS-CoV-2 pneumonia

Purpose

Comparison of deep learning algorithm, radiomics and subjective assessment of chest CT for predicting outcome (death or recovery) and intensive care unit (ICU) admission in patients with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection.

Methods

The multicenter, ethical committee-approved, retrospective study included non-contrast-enhanced chest CT of 221 SARS-CoV-2 positive patients from Italy (n = 196 patients; mean age 64 ± 16 years) and Denmark (n = 25; mean age 69 ± 13 years). A thoracic radiologist graded presence, type and extent of pulmonary opacities and severity of motion artifacts in each lung lobe on all chest CTs. Thin-section CT images were processed with CT Pneumonia Analysis Prototype (Siemens Healthineers) which yielded segmentation masks from a deep learning (DL) algorithm to derive features of lung abnormalities such as opacity scores, mean HU, as well as volume and percentage of all-attenuation and high-attenuation (opacities >−200 HU) opacities. Separately, whole lung radiomics were obtained for all CT exams. Analysis of variance and multiple logistic regression were performed for data analysis.

Results

Moderate to severe respiratory motion artifacts affected nearly one-quarter of chest CTs in patients. Subjective severity assessment, DL-based features and radiomics predicted patient outcome (AUC 0.76 vs AUC 0.88 vs AUC 0.83) and need for ICU admission (AUC 0.77 vs AUC 0.0.80 vs 0.82). Excluding chest CT with motion artifacts, the performance of DL-based and radiomics features improve for predicting ICU admission.

Conclusion

DL-based and radiomics features of pulmonary opacities from chest CT were superior to subjective assessment for differentiating patients with favorable and adverse outcomes.

Radiation Dose Reduction in Kidney Stone CT: A Randomized, Facility-Based Intervention

OBJECTIVE

Kidney stones are common, tend to recur, and afflict a young population. Despite evidence and recommendations, adoption of reduced-radiation dose CT (RDCT) for kidney stone CT (KSCT) is slow. We sought to design and test an intervention to improve adoption of RDCT protocols for KSCT using a randomized facility-based intervention.

METHODS

Facilities contributing at least 40 KSCTs to the American College of Radiology dose index registry (DIR) during calendar year 2015 were randomized to intervention or control groups. The Dose Optimization for Stone Evaluation intervention included customized CME modules, personalized consultation, and protocol recommendations for RDCT. Dose length product (DLP) of all KSCTs was recorded at baseline (2015) and compared with 2017, 2018, and 2019. Change in mean DLP was compared between facilities that participated (intervened-on), facilities randomized to intervention that did not participate (intervened-off), and control facilities. Difference-in-difference between intervened-on and control facilities is reported before and after intervention.

RESULTS

Of 314 eligible facilities, 155 were randomized to intervention and 159 to control. There were 25 intervened-on facilities, 71 intervened-off facilities, and 96 control facilities. From 2015 to 2017, there was a drop of 110 mGy ∙ cm (a 16% reduction) in the mean DLP in the intervened-on group, which was significantly lower compared with the control group (P < .05). The proportion of RDCTs increased for each year in the intervened-on group relative to the other groups for all 3 years (P < .01).

DISCUSSION

The Dose Optimization for Stone Evaluation intervention resulted in a significant (P < .05) and persistent reduction in mean radiation doses for engaged facilities performing KSCTs.

Five-star rating system for acceptable quality and dose in CT

KEY POINTS

• Recent papers have shown examples of the methodology involved in integrating image quality with radiation dose and assessing acceptable quality dose (AQD).• As a further step in that direction, translating a 5-point score into a 5-star rating shall be helpful for wider and uniform application and shall be in line with the popular use of the 5-star rating.

AI-based improvement in lung cancer detection on chest radiographs: results of a multi-reader study in NLST dataset

OBJECTIVE

Assess if deep learning-based artificial intelligence (AI) algorithm improves reader performance for lung cancer detection on chest X-rays (CXRs).

METHODS

This reader study included 173 images from cancer-positive patients (n = 98) and 346 images from cancer-negative patients (n = 196) selected from National Lung Screening Trial (NLST). Eight readers, including three radiology residents, and five board-certified radiologists, participated in the observer performance test. AI algorithm provided image-level probability of pulmonary nodule or mass on CXRs and a heatmap of detected lesions. Reader performance was compared with AUC, sensitivity, specificity, false-positives per image (FPPI), and rates of chest CT recommendations.

RESULTS

With AI, the average sensitivity of readers for the detection of visible lung cancer increased for residents, but was similar for radiologists compared to that without AI (0.61 [95% CI, 0.55-0.67] vs. 0.72 [95% CI, 0.66-0.77], p = 0.016 for residents, and 0.76 [95% CI, 0.72-0.81] vs. 0.76 [95% CI, 0.72-0.81, p = 1.00 for radiologists), while false-positive findings per image (FPPI) was similar for residents, but decreased for radiologists (0.15 [95% CI, 0.11-0.18] vs. 0.12 [95% CI, 0.09-0.16], p = 0.13 for residents, and 0.24 [95% CI, 0.20-0.29] vs. 0.17 [95% CI, 0.13-0.20], p < 0.001 for radiologists). With AI, the average rate of chest CT recommendation in patients positive for visible cancer increased for residents, but was similar for radiologists (54.7% [95% CI, 48.2-61.2%] vs. 70.2% [95% CI, 64.2-76.2%], p < 0.001 for residents and 72.5% [95% CI, 68.0-77.1%] vs. 73.9% [95% CI, 69.4-78.3%], p = 0.68 for radiologists), while that in cancer-negative patients was similar for residents, but decreased for radiologists (11.2% [95% CI, 9.6-13.1%] vs. 9.8% [95% CI, 8.0-11.6%], p = 0.32 for residents and 16.4% [95% CI, 14.7-18.2%] vs. 11.7% [95% CI, 10.2-13.3%], p < 0.001 for radiologists).

CONCLUSIONS

AI algorithm can enhance the performance of readers for the detection of lung cancers on chest radiographs when used as second reader.

KEY POINTS

• Reader study in the NLST dataset shows that AI algorithm had sensitivity benefit for residents and specificity benefit for radiologists for the detection of visible lung cancer. • With AI, radiology residents were able to recommend more chest CT examinations (54.7% vs 70.2%, p < 0.001) for patients with visible lung cancer. • With AI, radiologists recommended significantly less proportion of unnecessary chest CT examinations (16.4% vs. 11.7%, p < 0.001) in cancer-negative patients.

A multi-center study of COVID-19 patient prognosis using deep learning-based CT image analysis and electronic health records

PURPOSE

As of August 30th, there were in total 25.1 million confirmed cases and 845 thousand deaths caused by coronavirus disease of 2019 (COVID-19) worldwide. With overwhelming demands on medical resources, patient stratification based on their risks is essential. In this multi-center study, we built prognosis models to predict severity outcomes, combining patients' electronic health records (EHR), which included vital signs and laboratory data, with deep learning- and CT-based severity prediction.

METHOD

We first developed a CT segmentation network using datasets from multiple institutions worldwide. Two biomarkers were extracted from the CT images: total opacity ratio (TOR) and consolidation ratio (CR). After obtaining TOR and CR, further prognosis analysis was conducted on datasets from INSTITUTE-1, INSTITUTE-2 and INSTITUTE-3. For each data cohort, generalized linear model (GLM) was applied for prognosis prediction.

RESULTS

For the deep learning model, the correlation coefficient of the network prediction and manual segmentation was 0.755, 0.919, and 0.824 for the three cohorts, respectively. The AUC (95 % CI) of the final prognosis models was 0.85(0.77,0.92), 0.93(0.87,0.98), and 0.86(0.75,0.94) for INSTITUTE-1, INSTITUTE-2 and INSTITUTE-3 cohorts, respectively. Either TOR or CR exist in all three final prognosis models. Age, white blood cell (WBC), and platelet (PLT) were chosen predictors in two cohorts. Oxygen saturation (SpO2) was a chosen predictor in one cohort.

CONCLUSION

The developed deep learning method can segment lung infection regions. Prognosis results indicated that age, SpO2, CT biomarkers, PLT, and WBC were the most important prognostic predictors of COVID-19 in our prognosis model.

Deep learning predicts cardiovascular disease risks from lung cancer screening low dose computed tomography

Cancer patients have a higher risk of cardiovascular disease (CVD) mortality than the general population. Low dose computed tomography (LDCT) for lung cancer screening offers an opportunity for simultaneous CVD risk estimation in at-risk patients. Our deep learning CVD risk prediction model, trained with 30,286 LDCTs from the National Lung Cancer Screening Trial, achieves an area under the curve (AUC) of 0.871 on a separate test set of 2,085 subjects and identifies patients with high CVD mortality risks (AUC of 0.768). We validate our model against ECG-gated cardiac CT based markers, including coronary artery calcification (CAC) score, CAD-RADS score, and MESA 10-year risk score from an independent dataset of 335 subjects. Our work shows that, in high-risk patients, deep learning can convert LDCT for lung cancer screening into a dual-screening quantitative tool for CVD risk estimation.

Prediction of burden and management of renal calculi from whole kidney radiomics: a multicenter study

Purpose

To assess if autosegmentation-assisted radiomics can predict disease burden, hydronephrosis, and treatment strategies in patients with renal calculi.

Methods

The local ethical committee-approved, retrospective study included 202 adult patients (mean age: 53 ± 17 years; male: 103; female: 99) who underwent clinically indicated, non-contrast abdomen-pelvis CT for suspected or known renal calculi. All CT examinations were reviewed to determine the presence (n = 123 patients) or absence (n = 79) of renal calculi. On CT images with renal calculi, each kidney stone was annotated and measured (maximum dimension, Hounsfield unit (HU), and combined and dominant stone volumes) using a HU threshold-based segmentation. We recorded the presence of hydronephrosis, number of renal calculi, and treatment strategies. Deidentified CT images were processed with the radiomics prototype (Radiomics, Frontier, Siemens Healthineers), which automatically segmented each kidney to obtain 1690 first-, shape-, and higher-order radiomics. Data were analyzed using multiple logistic regression analysis with areas under the curve (AUC) as output.

Results

Among 202 patients, only 28 patients (18%) needed procedural treatment (lithotripsy or ureteroscopic stone extraction). Gray-level co-occurrence matrix (GLCM) and gray-level run length matrix (GLRLM) differentiated patients with and without procedural treatment (AUC 0.91, 95% CI 0.85–0.92). Higher-order radiomics (gray-level size zone matrix – GLSZM) differentiated kidneys with and without hydronephrosis (AUC: 0.99, p < 0.001) as well those with different stone volumes (AUC up to 0.89, 95% CI 0.89–0.92).

Conclusion

Automated segmentation and radiomics of entire kidneys can assess hydronephrosis presence, stone burden, and treatment strategies for renal calculi with AUCs > 0.85.

Investigating centering, scan length, and arm position impact on radiation dose across 4 countries from 4 continents during pandemic: Mitigating key radioprotection issues

Purpose

Optimization of CT scan practices can help achieve and maintain optimal radiation protection. The aim was to assess centering, scan length, and positioning of patients undergoing chest CT for suspected or known COVID-19 pneumonia and to investigate their effect on associated radiation doses.

Methods

With respective approvals from institutional review boards, we compiled CT imaging and radiation dose data from four hospitals belonging to four countries (Brazil, Iran, Italy, and USA) on 400 adult patients who underwent chest CT for suspected or known COVID-19 pneumonia between April 2020 and August 2020. We recorded patient demographics and volume CT dose index (CTDI_vol) and dose length product (DLP). From thin-section CT images of each patient, we estimated the scan length and recorded the first and last vertebral bodies at the scan start and end locations. Patient mis-centering and arm position were recorded. Data were analyzed with analysis of variance (ANOVA).

Results

The extent and frequency of patient mis-centering did not differ across the four CT facilities (>0.09). The frequency of patients scanned with arms by their side (11–40% relative to those with arms up) had greater mis-centering and higher CTDI_vol and DLP at 2/4 facilities (p = 0.027–0.05). Despite lack of variations in effective diameters (p = 0.14), there were significantly variations in scan lengths, CTDI_vol and DLP across the four facilities (p < 0.001).

Conclusions

Mis-centering, over-scanning, and arms by the side are frequent issues with use of chest CT in COVID-19 pneumonia and are associated with higher radiation doses.

Artificial intelligence-based vessel suppression for detection of sub-solid nodules in lung cancer screening computed tomography

Background

Lung cancer screening (LCS) with low-dose computed tomography (LDCT) helps early lung cancer detection, commonly presenting as small pulmonary nodules. Artificial intelligence (AI)-based vessel suppression (AI-VS) and automatic detection (AI-AD) algorithm can improve detection of subsolid nodules (SSNs) on LDCT. We assessed the impact of AI-VS and AI-AD in detection and classification of SSNs [ground-glass nodules (GGNs) and part-solid nodules (PSNs)], on LDCT performed for LCS.

Methods

Following regulatory approval, 123 LDCT examinations with sub-solid pulmonary nodules (average diameter ≥6 mm) were processed to generate three image series for each examination-unprocessed, AI-VS, and AI-AD series with annotated lung nodules. Two thoracic radiologists in consensus formed the standard of reference (SOR) for this study. Two other thoracic radiologists (R1 and R2; 5 and 10 years of experience in thoracic CT image interpretation) independently assessed the unprocessed images alone, then together with AI-VS series, and finally with AI-AD for detecting all ≥6 mm GGN and PSN. We performed receiver operator characteristics (ROC) and Cohen's Kappa analyses for statistical analyses.

Results

On unprocessed images, R1 and R2 detected 232/310 nodules (R1: 114 GGN, 118 PSN) and 255/310 nodules (R2: 122 GGN, 133 PSN), respectively (P>0.05). On AI-VS images, they detected 249/310 nodules (119 GGN, 130 PSN) and 277/310 nodules (128 GGN, 149 PSN), respectively (P≥0.12). When compared to the SOR, accuracy (AUC) for detection of PSN on the AI-VS images (AUC 0.80-0.81) was greater than on the unprocessed images (AUC 0.70-0.76). AI-VS images enabled detection of solid components in five nodules deemed as GGN on the unprocessed images. Accuracy of AI-AD was lower than both the radiologists (AUC 0.60-0.72).

Conclusions

AI-VS improved the detection and classification of SSN into GGN and PSN on LDCT of the chest for the two radiologist (R1 and R2) readers.