Comparison of emergency cranial CT interpretation between radiology residents and neuroradiologists: transverse versus three-dimensional images

The use of individualized 3D-printed models on trainee and patient education, and surgical planning for robotic partial nephrectomies

3D printing is a growing tool in surgical education to visualize and teach complex procedures. Previous studies demonstrating the usefulness of 3D models as teaching tools for partial nephrectomy used highly detailed models costing between $250 and 1000. We aimed to create thorough, inexpensive 3D models to accelerate learning for trainees and increase health literacy in patients. Patient-specific, cost-effective ($30-50) 3D models of the affected urologic structures were created using pre-operative imaging of 40 patients undergoing partial nephrectomy at Thomas Jefferson University Hospital (TJUH) between July 2020 and May 2021. Patients undergoing surgery filled out a survey before and after seeing the model to assess patient understanding of their kidney, pathophysiology, surgical procedure, and risks of surgery. Three urological residents, one fellow, and six attendings filled out separate surveys to assess their surgical plan and confidence before and after seeing the model. In a third survey, they ranked how much the model helped their comprehension and confidence during surgery. Patient understanding of all four subjects significantly improved after seeing the 3D model (P < 0.001). The urology residents (P < 0.001) and fellow (P < 0.001) reported significantly increased self-confidence after interacting with the model. Attending surgeon confidence increased significantly after seeing the 3D model (P < 0.01) as well. Cost-effective 3D models are effective learning tools and assist with the evaluation of patients presenting with renal masses, and increase patient, resident, and fellow understanding in partial nephrectomies. Further research should continue to explore the utility of inexpensive models in other urologic procedures.

Impact of lung-RADS classification system on the accurate diagnosis of pulmonary nodular lesions in oncology patients

Background

Lung assessment is highly recommended in the management of oncology patients as it is the commonest affected site in metastatic dissemination. The low-dose CT with nodule reporting system based on Lung Reporting and Data System (lung-RADS) is a promising non-invasive tool for the characterization of incidentally detected pulmonary nodules. The authors aimed to assess the accuracy of the “lung-RADS” classification system as a non-invasive tool for the characterization of any newly developed pulmonary nodules among oncology patients. Ethics committee approval and informed written consent were obtained from the studied patients. A non-contrast LDCT study was performed on all patients with a nodule reporting system based on the lung-RADS classification system applied for evaluation of each detected pulmonary nodule. Diagnoses were established using the help of either histopathology or follow-up clinical results as a gold standard.

Results

In this prospective study, we enrolled 187 known malignancy patients with 200 suspicious newly developed pulmonary nodules. Their mean patient age was 48.4 ± 9.7 years. The studied 200 pulmonary nodular lesions were categorized using a nodule reporting system based on the lung-RADS into 6 sub-groups with 122 lesions found to be malignant and 78 lesions were of benign etiology, which showed a high sensitivity of 92.08%, specificity of 78.79%, and accuracy of 85.50% with 81.58% positive predictive value and 90.70% negative predictive value in the diagnosis of pulmonary nodules in cancer patients.

Conclusion

Low-density CT with a nodule reporting system based on the lung-RADS classification system was found to be an accurate non-invasive tool to characterize and to risk stratify pulmonary nodules in oncology patients.

Differentiation of adenocarcinoma in situ with alveolar collapse from minimally invasive adenocarcinoma or invasive adenocarcinoma appearing as part-solid ground-glass nodules (≤ 2 cm) using computed tomography

PURPOSE

To investigate the differentiating computed tomographic (CT) features between adenocarcinoma in situ (AIS) with alveolar collapse and minimally invasive adenocarcinoma (MIA) or invasive adenocarcinoma (IA) appearing as part-solid nodules.

METHODS

A total of 147 consecutive patients with 157 pathology-confirmed part-solid ground-glass nodules (GGNs) ≤ 20 mm without other pathological condition such as inflammation and fibrosis who underwent chest CT were included.

RESULTS

The 157 part-solid GGNs included 33 (21.02%) pathologically confirmed AISs with alveolar collapse. Multivariate analysis revealed that smaller lesion size (odds ratio [OR] 0.671), and well-defined border (OR 5.544), concentrated distribution (OR 7.994), and homogeneity of the solid portion (OR 4.365) were significant independent predictors for differentiating AIS with alveolar collapse from MIA (P < 0.05) with excellent accuracy (area under receiver operating characteristic [ROC] curve, 0.902). Multivariate analysis revealed that smaller lesion size (OR 0.782), and size (OR 0.821), well-defined border (OR 5.752), and homogeneity of solid portion (OR 6.182) were significant independent predictors differentiating AIS with alveolar collapse from IA (P < 0.05) with excellent accuracy (area under ROC curve 0.910).

CONCLUSION

Among part-solid GGNs, AIS with alveolar collapse can be accurately differentiated from MIA on the basis of smaller lesion size, well-defined border, concentrated distribution, and homogeneity of solid portion, and from IA according to smaller lesion size, and smaller size, well-defined border, and homogeneity of solid portion.

Human-recognizable CT image features of subsolid lung nodules associated with diagnosis and classification by convolutional neural networks

OBJECTIVES

The interpretability of convolutional neural networks (CNNs) for classifying subsolid nodules (SSNs) is insufficient for clinicians. Our purpose was to develop CNN models to classify SSNs on CT images and to investigate image features associated with the CNN classification.

METHODS

CT images containing SSNs with a diameter of ≤ 3 cm were retrospectively collected. We trained and validated CNNs by a 5-fold cross-validation method for classifying SSNs into three categories (benign and preinvasive lesions [PL], minimally invasive adenocarcinoma [MIA], and invasive adenocarcinoma [IA]) that were histologically confirmed or followed up for 6.4 years. The mechanism of CNNs on human-recognizable CT image features was investigated and visualized by gradient-weighted class activation map (Grad-CAM), separated activation channels and areas, and DeepDream algorithm.

RESULTS

The accuracy was 93% for classifying 586 SSNs from 569 patients into three categories (346 benign and PL, 144 MIA, and 96 IA in 5-fold cross-validation). The Grad-CAM successfully located the entire region of image features that determined the final classification. Activated areas in the benign and PL group were primarily smooth margins (p < 0.001) and ground-glass components (p = 0.033), whereas in the IA group, the activated areas were mainly part-solid (p < 0.001) and solid components (p < 0.001), lobulated shapes (p < 0.001), and air bronchograms (p < 0.001). However, the activated areas for MIA were variable. The DeepDream algorithm showed the image features in a human-recognizable pattern that the CNN learned from a training dataset.

CONCLUSION

This study provides medical evidence to interpret the mechanism of CNNs that helps support the clinical application of artificial intelligence.

KEY POINTS

• CNN achieved high accuracy (93%) in classifying subsolid nodules on CT images into three categories: benign and preinvasive lesions, MIA, and IA. • The gradient-weighted class activation map (Grad-CAM) located the entire region of image features that determined the final classification, and the visualization of the separated activated areas was consistent with radiologists' expertise for diagnosing subsolid nodules. • DeepDream showed the image features that CNN learned from a training dataset in a human-recognizable pattern.

Comparison of physical image qualities and artifact indices for head computed tomography in the axial and helical scan modes

This study aimed to validate the clinically demonstrated equivalency of the axial and helical scan modes (AS and HS, respectively) for head computed tomography (CT) using physical image quality measures and artifact indices (AIs). Two 64-row multi-detector row CT systems (CT-A and CT-B) were used for comparing AS and HSs with detector rows of 64 and 32. The modulation transfer function (MTF), noise power spectrum (NPS), and slice sensitivity profile were measured using a CT dose index corresponding to clinical use. The system performance function (SPF) was calculated as MTF²/NPS. The AI of streak artifacts in the skull base was measured using an image obtained of a head phantom, while the AI of motion artifacts was measured from images obtained during the head phantom was in motion. For CT-A, the 50%MTFs were 7% to 9% higher in the HS than the AS, and the higher MTFs of HS associated NPS increases. For CT-B, the MTFs and NPSs were almost equivalent between the AS and HS, respectively. Consequently, the SPFs of AS and HS were nearly identical for both CT systems. For both CT systems, the skull base AI did not differ significantly between AS and HS, while the motion AIs of HS were significantly better than of AS. The superior motion AI in the HS indicated the effectiveness of HS on moving patients.

Cortical Changes in Epilepsy Patients With Focal Cortical Dysplasia: New Insights With T2 Mapping

BACKGROUND

In epilepsy patients with focal cortical dysplasia (FCD) as the epileptogenic focus, global cortical signal changes are generally not visible on conventional MRI. However, epileptic seizures or antiepileptic medication might affect normal-appearing cerebral cortex and lead to subtle damage.

PURPOSE

To investigate cortical properties outside FCD regions with T₂ -relaxometry.

STUDY TYPE

Prospective study.

SUBJECTS

Sixteen patients with epilepsy and FCD and 16 age-/sex-matched healthy controls.

FIELD STRENGTH/SEQUENCE

3T, fast spin-echo T₂ -mapping, fluid-attenuated inversion recovery (FLAIR), and synthetic T₁ -weighted magnetization-prepared rapid acquisition of gradient-echoes (MP-RAGE) datasets derived from T₁ -maps.

ASSESSMENT

Reconstruction of the white matter and cortical surfaces based on MP-RAGE structural images was performed to extract cortical T₂ values, excluding lesion areas. Three independent raters confirmed that morphological cortical/juxtacortical changes in the conventional FLAIR datasets outside the FCD areas were definitely absent for all patients. Averaged global cortical T₂ values were compared between groups. Furthermore, group comparisons of regional cortical T₂ values were performed using a surface-based approach. Tests for correlations with clinical parameters were carried out.

STATISTICAL TESTS

General linear model analysis, permutation simulations, paired and unpaired t-tests, and Pearson correlations.

RESULTS

Cortical T₂ values were increased outside FCD regions in patients (83.4 ± 2.1 msec, control group 81.4 ± 2.1 msec, P = 0.01). T₂ increases were widespread, affecting mainly frontal, but also parietal and temporal regions of both hemispheres. Significant correlations were not observed (P ≥ 0.55) between cortical T₂ values in the patient group and the number of seizures in the last 3 months or the number of anticonvulsive drugs in the medical history.

DATA CONCLUSION

Widespread increases in cortical T₂ in FCD-associated epilepsy patients were found, suggesting that structural epilepsy in patients with FCD is not only a symptom of a focal cerebral lesion, but also leads to global cortical damage not visible on conventional MRI.

EVIDENCE LEVEL

21 TECHNICAL EFFICACY STAGE: 3 J. MAGN. RESON. IMAGING 2020;52:1783-1789.

Precise diagnosis of intracranial hemorrhage and subtypes using a three-dimensional joint convolutional and recurrent neural network

OBJECTIVES

To evaluate the performance of a novel three-dimensional (3D) joint convolutional and recurrent neural network (CNN-RNN) for the detection of intracranial hemorrhage (ICH) and its five subtypes (cerebral parenchymal, intraventricular, subdural, epidural, and subarachnoid) in non-contrast head CT.

METHODS

A total of 2836 subjects (ICH/normal, 1836/1000) from three institutions were included in this ethically approved retrospective study, with a total of 76,621 slices from non-contrast head CT scans. ICH and its five subtypes were annotated by three independent experienced radiologists, with majority voting as reference standard for both the subject level and the slice level. Ninety percent of data was used for training and validation, and the rest 10% for final evaluation. A joint CNN-RNN classification framework was proposed, with the flexibility to train when subject-level or slice-level labels are available. The predictions were compared with the interpretations from three junior radiology trainees and an additional senior radiologist.

RESULTS

It took our algorithm less than 30 s on average to process a 3D CT scan. For the two-type classification task (predicting bleeding or not), our algorithm achieved excellent values (≥ 0.98) across all reporting metrics on the subject level. For the five-type classification task (predicting five subtypes), our algorithm achieved > 0.8 AUC across all subtypes. The performance of our algorithm was generally superior to the average performance of the junior radiology trainees for both two-type and five-type classification tasks.

CONCLUSIONS

The proposed method was able to accurately detect ICH and its subtypes with fast speed, suggesting its potential for assisting radiologists and physicians in their clinical diagnosis workflow.

KEY POINTS

• A 3D joint CNN-RNN deep learning framework was developed for ICH detection and subtype classification, which has the flexibility to train with either subject-level labels or slice-level labels. • This deep learning framework is fast and accurate at detecting ICH and its subtypes. • The performance of the automated algorithm was superior to the average performance of three junior radiology trainees in this work, suggesting its potential to reduce initial misinterpretations.

Virtual Monoenergetic Images from Spectral Detector CT Enable Radiation Dose Reduction in Unenhanced Cranial CT

BACKGROUND AND PURPOSE

Our aim was to evaluate whether improved gray-white matter differentiation in cranial CT by means of 65- keV virtual monoenergetic images enables a radiation dose reduction compared to conventional images.

MATERIALS AND METHODS

One hundred forty consecutive patients undergoing 171 spectral detector CTs of the head between February and November 2017 (56 ± 19 years of age; male/female ratio, 56%/44%) were retrospectively included. The tube current-time product was reduced during the study period, resulting in 61, 55, and 55 patients being examined with 320, 290, and 260 mAs, respectively. All other scanning parameters were kept identical. The volume CT dose index was recorded. ROIs were placed in gray and white matter on conventional images and copied to identical positions in 65- keV virtual monoenergetic images. The contrast-to-noise ratio was calculated. Two radiologists blinded to the reconstruction technique evaluated image quality on a 5-point Likert-scale. Statistical assessment was performed using ANOVA and Wilcoxon test adjusted for multiple comparisons.

RESULTS

The mean volume CT dose index was 55, 49.8, and 44.7 mGy using 320, 290, and 260 mAs, respectively. Irrespective of the volume CT dose index, noise was significantly lower in 65- keV virtual monoenergetic images compared with conventional images (65- keV virtual monoenergetic images/conventional images: extraocular muscle with 49.8 mGy, 3.7 ± 1.3/5.6 ± 1.6 HU, P < .001). Noise slightly increased with a reduced radiation dose (eg, extraocular muscle in conventional images: 5.3 ± 1.4/5.6 ± 1.6/6.1 ± 2.1 HU). Overall, the contrast-to-noise ratio in 65- keV virtual monoenergetic images was superior to that in conventional images irrespective of the volume CT dose index (P < .001). Particularly, 65-keV virtual monoenergetic images with 44.7 mGy showed significantly lower noise and a higher contrast-to-noise ratio than conventional images with 55 mGy (P < .001). Subjective analysis confirmed better image quality in 65- keV virtual monoenergetic images, even using 44.7 mGy.

CONCLUSIONS

The 65-keV virtual monoenergetic images from spectral detector CT allow radiation dose reduction in cranial CT. While this proof of concept included a radiation dose reduction of 19%, our data suggest that even greater reduction appears achievable.

Validation of an informatics tool to assess resident's progress in developing reporting skills

Background

Diagnostic radiology residency programs pursuits as main objectives of the development of diagnostic capabilities and written communication skills to answer clinicians’ questions of referring clinicians.

There has been also an increasing focus on competencies, rather than just education inputs. Then, to show ongoing professional development is necessary for a system to assess and document resident’s competence in these areas.

Therefore, we propose the implementation of an informatics tool to objectively assess resident’s progress in developing diagnostics and reporting skills. We expect to found decreased preliminary report-final report variability within the course of each year of the residency program.

Results

We analyzed 12,162 evaluations from 32 residents (8 residents per year in a 4-year residency program) in a 7-month period. 73.96% of these evaluations belong to 2nd-year residents.

We chose two indicators to study the evolution of evaluations: the total of discrepancies over the total of preliminary reports (excluding score 0) and the total of likely to be clinically significant discrepancies (scores 2b, 3b, and 4b) over the total of preliminary reports (excluding score 0).

With the analysis of these two indicators over the evaluations of 2nd-year residents, we found a slight decrease in the value of the first indicator and relative stable behavior of the second one.

Conclusions

This tool is useful for objective assessment of reporting skill of radiology residents. It can provide an opportunity for continuing medical education with case-based learning from those cases with clinically significant discrepancies between the preliminary and the final report.

Electronic supplementary material

The online version of this article (10.1186/s13244-019-0772-0) contains supplementary material, which is available to authorized users.

Histogram-based models on non-thin section chest CT predict invasiveness of primary lung adenocarcinoma subsolid nodules

109 pathologically proven subsolid nodules (SSN) were segmented by 2 readers on non-thin section chest CT with a lung nodule analysis software followed by extraction of CT attenuation histogram and geometric features. Functional data analysis of histograms provided data driven features (FPC1,2,3) used in further model building. Nodules were classified as pre-invasive (P1, atypical adenomatous hyperplasia and adenocarcinoma in situ), minimally invasive (P2) and invasive adenocarcinomas (P3). P1 and P2 were grouped together (T1) versus P3 (T2). Various combinations of features were compared in predictive models for binary nodule classification (T1/T2), using multiple logistic regression and non-linear classifiers. Area under ROC curve (AUC) was used as diagnostic performance criteria. Inter-reader variability was assessed using Cohen’s Kappa and intra-class coefficient (ICC). Three models predicting invasiveness of SSN were selected based on AUC. First model included 87.5 percentile of CT lesion attenuation (Q.875), interquartile range (IQR), volume and maximum/minimum diameter ratio (AUC:0.89, 95%CI:[0.75 1]). Second model included FPC1, volume and diameter ratio (AUC:0.91, 95%CI:[0.77 1]). Third model included FPC1, FPC2 and volume (AUC:0.89, 95%CI:[0.73 1]). Inter-reader variability was excellent (Kappa:0.95, ICC:0.98). Parsimonious models using histogram and geometric features differentiated invasive from minimally invasive/pre-invasive SSN with good predictive performance in non-thin section CT.

A simple prediction model using size measures for discrimination of invasive adenocarcinomas among incidental pulmonary subsolid nodules considered for resection

OBJECTIVES

To develop and validate a concise prediction model using simple size measures for the discrimination of invasive pulmonary adenocarcinomas (IPAs) among incidentally detected subsolid nodules (SSNs) considered for resection and to compare its diagnostic performance with the Brock model.

METHODS

This retrospective institutional review board-approved study included 427 surgically resected SSNs (121 preinvasive lesions/minimally invasive adenocarcinomas [MIAs] and 306 IPAs) from 407 patients. After stratified random splitting of the study population into the training and validation sets (3:1), a simple logistic model was constructed using nodule size, solid proportion, and type for the differentiation of IPAs. Diagnostic performance of this model was compared with the original and modified Brock models using the DeLong method for area under the receiver-operating characteristic curve (AUC) and McNemar test for diagnostic sensitivity and specificity.

RESULTS

Our proposed model had an AUC of 0.859 in the validation set, while the original Brock model showed an AUC of 0.775 (p = 0.035) and the modified Brock model exhibited an AUC of 0.787 (p = 0.006). At equally high specificity of 90%, our proposed model exhibited significantly higher sensitivity (65.8%) than the original and modified Brock models (38.2% and 50.0%; p < 0.001 and 0.008, respectively).

CONCLUSIONS

Our study results demonstrated that the proposed concise model outperformed both Brock models, demonstrating its potential to be utilized as a specific tool to differentiate IPAs from preinvasive lesions and MIAs, which were considered for resection. External validation studies are warranted for the population with incidentally detected SSNs including small SSNs to confirm our observations.

KEY POINTS

• Size measures provided sufficient information for the risk stratification of surgical candidate incidental subsolid nodules. • Our proposed concise model showed higher diagnostic performance than the Brock model for incidentally detected subsolid nodules. • Our proposed model can specifically differentiate invasive adenocarcinomas among incidentally detected subsolid nodules and reduce overtreatment for indolent subsolid nodules.

Diameter of the Solid Component in Subsolid Nodules on Low-Dose Unenhanced Chest Computed Tomography: Measurement Accuracy for the Prediction of Invasive Component in Lung Adenocarcinoma

Objective

To determine if measurement of the diameter of the solid component in subsolid nodules (SSNs) on low-dose unenhanced chest computed tomography (CT) is as accurate as on standard-dose enhanced CT in prediction of pathological size of invasive component of lung adenocarcinoma.

Materials and Methods

From February 2012 to October 2015, 114 SSNs were identified in 105 patients that underwent low-dose unenhanced and standard-dose enhanced CT pre-operatively. Three radiologists independently measured the largest diameter of the solid component. Intraclass correlation coefficients (ICCs) were used to assess inter-reader agreement. We estimated measurement differences between the size of solid component and that of invasive component. We measured diagnostic accuracy of the prediction of invasive adenocarcinoma using a size criterion of a solid component ≥ 6 mm, and compared them using a generalized linear mixed model.

Results

Inter-reader agreement was excellent (ICC, 0.84.0.89). The mean ± standard deviation of absolute measurement differences between the solid component and invasive component was 4 ± 4 mm in low-dose unenhanced CT and 5 ± 4 mm in standard-dose enhanced CT. Diagnostic accuracy was 81.3% (95% confidence interval, 76.7.85.3%) in low-dose unenhanced CT and 76.6% (71.8.81.0%) in standard-dose enhanced CT, with no statistically significant difference (p = 0.130).

Conclusion

Measurement of the diameter of the solid component of SSNs on low-dose unenhanced chest CT was as accurate as on standard-dose enhanced CT for predicting the invasive component. Thus, low-dose unenhanced CT may be used safely in the evaluation of patients with SSNs.

Role of imaging in diagnosis, staging and follow-up of lung cancer

PURPOSE OF REVIEW

Primary lung cancer is still the number one cause of cancer death worldwide. Screening, detection and staging of lung cancer are important because the only potentially curative therapy today is surgical resection of early-stage lung cancer.

RECENT FINDINGS

Different imaging techniques can be used in these different processes. Recent advances in computed tomography (CT) technology have allowed investigation of novel methods for the evaluation of lung cancer. Recent advances in magnetic resonance technology and administration of contrast media have further improved the image quality and diagnostic capability of magnetic resonance. Positron emission tomography (PET)/CT has been shown to be superior to stand-alone PET or CT in the evaluation of lymph nodes and in the detection of distant metastases.

SUMMARY

The current recommended imaging required for lung cancer staging is CT of the thorax and PET/CT from skull base to mid-thigh. However, with the recent developments in the armamentarium of imaging techniques, the choice of one of these techniques can be directed by the presence of a technique in a local hospital and/or by the presence of an experienced person at that time.