Less-experienced Radiologists Research Articles

Diagnostic performance and image quality of an image-based denoising algorithm applied to radiation dose-reduced CT in diagnosing acute appendicitis.

To evaluate diagnostic performance and image quality of ultralow-dose CT (ULDCT) in diagnosing acute appendicitis with an image-based deep-learning denoising algorithm (IDLDA). This retrospective multicenter study included 180 patients (mean ± standard deviation, 29 ± 9 years; 91 female) who underwent contrast-enhanced 2-mSv CT for suspected appendicitis from February 2014 to August 2016. We simulated ULDCT from 2-mSv CT, reducing the dose by at least 50%. Then we applied an IDLDA on ULDCT to produce denoised ULDCT (D-ULDCT). Six radiologists with different experience levels (three board-certified radiologists and three residents) independently reviewed the ULDCT and D-ULDCT. They rated the likelihood of appendicitis and subjective image qualities (subjective image noise, diagnostic acceptability, and artificial sensation). One radiologist measured image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). We used the receiver operating characteristic (ROC) analyses, Wilcoxon's signed-rank tests, and paired t-tests. The area under the ROC curves (AUC) for diagnosing appendicitis ranged 0.90-0.97 for ULDCT and 0.94-0.97 for D-ULDCT. The AUCs of two residents were significantly higher on D-ULDCT (AUC difference = 0.06 [95% confidence interval, 0.01-0.11; p = .022] and 0.05 [0.00-0.10; p = .046], respectively). D-ULDCT provided better subjective image noise and diagnostic acceptability to all six readers. However, the response of board-certified radiologists and residents differed in artificial sensation (all p ≤ .003). D-ULDCT showed significantly lower image noise, higher SNR, and higher CNR (all p < .001). An IDLDA can provide better ULDCT image quality and enhance diagnostic performance for less-experienced radiologists.

Diagnosis of skull-base invasion by nasopharyngeal tumors on CT with a deep-learning approach

PurposeTo develop a convolutional neural network (CNN) model to diagnose skull-base invasion by nasopharyngeal malignancies in CT images and evaluate the model’s diagnostic performance.Materials and methodsWe divided 100 malignant nasopharyngeal tumor lesions into a training (n = 70) and a test (n = 30) dataset. Two head/neck radiologists reviewed CT and MRI images and determined the positive/negative skull-base invasion status of each case (training dataset: 29 invasion-positive and 41 invasion-negative; test dataset: 13 invasion-positive and 17 invasion-negative). Preprocessing involved extracting continuous slices of the nasopharynx and clivus. The preprocessed training dataset was used for transfer learning with Residual Neural Networks 50 to create a diagnostic CNN model, which was then tested on the preprocessed test dataset to determine the invasion status and model performance. Original CT images from the test dataset were reviewed by a radiologist with extensive head/neck imaging experience (senior reader: SR) and another less-experienced radiologist (junior reader: JR). Gradient-weighted class activation maps (Grad-CAMs) were created to visualize the explainability of the invasion status classification.ResultsThe CNN model’s diagnostic accuracy was 0.973, significantly higher than those of the two radiologists (SR: 0.838; JR: 0.595). Receiver operating characteristic curve analysis gave an area under the curve of 0.953 for the CNN model (versus 0.832 and 0.617 for SR and JR; both p < 0.05). The Grad-CAMs suggested that the invasion-negative cases were present predominantly in bone marrow, while the invasion-positive cases exhibited osteosclerosis and nasopharyngeal masses.ConclusionsThis CNN technique would be useful for CT-based diagnosis of skull-base invasion by nasopharyngeal malignancies.

Clinical Application of Computer-Aided Diagnosis System in Breast Ultrasound: A Prospective Multicenter Study.

Ultrasound tends to present very high sensitivity but relatively low specificity and positive predictive value (PPV), which would result in unnecessary breast biopsies. The purpose of this study is to analyze the diagnostic performance of computer-aided diagnosis (CAD) (S-Detect) system in differentiating breast lesions and reducing unnecessary biopsies in non-university hospitals in less-developed regions of China. The study was a prospective multicenter study from 8 hospitals. The ultrasound images, and cine, CAD analysis, and BI-RADS were recorded. The accuracy, sensitivity, specificity, PPV, negative predictive value (NPV), and area under the curve (AUC) were analyzed and compared between CAD and radiologists. The Youden Index (YI) was used to determine optimal cut-off for the number of planes to downgrade. A total of 491 breast lesions were included in the study. Less-experienced radiologists combined CAD was superior to less-experienced radiologists alone in AUC (0.878 vs 0.712, p < 0.001), and specificity (81.3% vs 44.6%, p < 0.001). There was no statistical difference in AUC (0.891 vs 0.878, p = 0.346), and specificity (82.3% vs 81.3%, p = 0.791) between experienced radiologists and less-experienced radiologists combined CAD. With CAD assistance, the biopsy rate of less-experienced radiologists was significantly decreased (100.0% vs 25.6%, p < 0.001), and malignant rate of biopsy was significantly increased (15.0% vs 43.9%, p < 0.001). CAD system can be an effective auxiliary tool in differentiating breast lesions and reducing unnecessary biopsies for radiologists from non-university hospitals in less-developed regions of China.

Uterine mesenchymal tumors: development and preliminary results of a magnetic resonance imaging (MRI) diagnostic algorithm.

The aim of our study is to propose a diagnostic algorithm to guide MRI findings interpretation and malignancy risk stratification of uterine mesenchymal masses with a multiparametric step-by-step approach. A non-interventional retrospective multicenter study was performed: Preoperative MRI of 54 uterine masses was retrospectively evaluated. Firstly, the performance of MRI with monoparametric and multiparametric approach was assessed. Reference standard for final diagnosis was surgical pathologic result (n = 53 patients) or at least 1-year MR imaging follow-up (n = 1 patient). Subsequently, a diagnostic algorithm was developed for MR interpretation, resulting in a Likert score from 1 to 5 predicting risk of malignancy of the uterine lesion. The accuracy and reproducibility of the MRI scoring system were then tested: 26 preoperative pelvic MRI were double-blind evaluated by a senior (SR) and junior radiologist (JR). Diagnostic performances and the agreement between the two readers with and without the application of the proposed algorithm were compared, using histological results as standard reference. Multiparametric approach showed the best diagnostic performance in terms of accuracy (94.44%,) and specificity (97.56%). DWI was confirmed as the most sensible parameter with a relative high specificity: low ADC values (mean 0.66) significantly correlated to uterine sarcomas diagnosis (p < 0.01). Proposed algorithm allowed to improve both JR and SR performance (algorithm-aided accuracy 88.46% and 96%, respectively) and determined a significant increase in inter-observer agreement, helping even the less-experienced radiologist in this difficult differential diagnosis. Uterine leiomyomas and sarcomas often show an overlap of clinical and imaging features. The application of a diagnostic algorithm can help radiologists to standardize their approach to a complex myometrial mass and to easily identify suspicious MRI features favoring malignancy.

Machine learning combined with radiomics and deep learning features extracted from CT images: a novel AI model to distinguish benign from malignant ovarian tumors

BackgroundTo develop an artificial intelligence (AI) model with radiomics and deep learning (DL) features extracted from CT images to distinguish benign from malignant ovarian tumors.MethodsWe enrolled 149 patients with pathologically confirmed ovarian tumors. A total of 185 tumors were included and divided into training and testing sets in a 7:3 ratio. All tumors were manually segmented from preoperative contrast-enhanced CT images. CT image features were extracted using radiomics and DL. Five models with different combinations of feature sets were built. Benign and malignant tumors were classified using machine learning (ML) classifiers. The model performance was compared with five radiologists on the testing set.Results Among the five models, the best performing model is the ensemble model with a combination of radiomics, DL, and clinical feature sets. The model achieved an accuracy of 82%, specificity of 89% and sensitivity of 68%. Compared with junior radiologists averaged results, the model had a higher accuracy (82% vs 66%) and specificity (89% vs 65%) with comparable sensitivity (68% vs 67%). With the assistance of the model, the junior radiologists achieved a higher average accuracy (81% vs 66%), specificity (80% vs 65%), and sensitivity (82% vs 67%), approaching to the performance of senior radiologists.Conclusions We developed a CT-based AI model that can differentiate benign and malignant ovarian tumors with high accuracy and specificity. This model significantly improved the performance of less-experienced radiologists in ovarian tumor assessment, and may potentially guide gynecologists to provide better therapeutic strategies for these patients.

Clinical value of artificial intelligence in thyroid ultrasound: a prospective study from the real world.

To evaluate the diagnostic performance of a commercial artificial intelligence (AI)-assisted ultrasonography (US) for thyroid nodules and to validate its value in real-world medical practice. From March 2021 to July 2021, 236 consecutive patients with 312 suspicious thyroid nodules were prospectively enrolled in this study. One experienced radiologist performed US examinations with a real-time AI system (S-Detect). US images and AI reports of the nodules were recorded. Nine residents and three senior radiologists were invited to make a "benign" or "malignant" diagnosis based on recorded US images without knowing the AI reports. After referring to AI reports, the diagnosis was made again. The diagnostic performance of AI, residents, and senior radiologists with and without AI reports were analyzed. The sensitivity, accuracy, and AUC of the AI system were 0.95, 0.84, and 0.753, respectively, and were not statistically different from those of the experienced radiologists, but were superior to those of the residents (all p < 0.01). The AI-assisted resident strategy significantly improved the accuracy and sensitivity for nodules ≤ 1.5 cm (all p < 0.01), while reducing the unnecessary biopsy rate by up to 27.7% for nodules > 1.5 cm (p = 0.034). The AI system achieved performance, for cancer diagnosis, comparable to that of an average senior thyroid radiologist. The AI-assisted strategy can significantly improve the overall diagnostic performance for less-experienced radiologists, while increasing the discovery of thyroid cancer ≤ 1.5 cm and reducing unnecessary biopsies for nodules > 1.5 cm in real-world medical practice. • The AI system reached a senior radiologist-like level in the evaluation of thyroid cancer and could significantly improve the overall diagnostic performance of residents. • The AI-assisted strategy significantly improved ≤ 1.5 cm thyroid cancer screening AUC, accuracy, and sensitivity of the residents, leading to an increased detection of thyroid cancer while maintaining a comparable specificity to that of radiologists alone. • The AI-assisted strategy significantly reduced the unnecessary biopsy rate for thyroid nodules > 1.5 cm by the residents, while maintaining a comparable sensitivity to that of radiologists alone.

A concurrent, deep learning-based computer-aided detection system for prostate multiparametric MRI: a performance study involving experienced and less-experienced radiologists.

To evaluate the effect of a deep learning-based computer-aided diagnosis (DL-CAD) system on experienced and less-experienced radiologists in reading prostate mpMRI. In this retrospective, multi-reader multi-case study, a consecutive set of 184 patients examined between 01/2018 and 08/2019 were enrolled. Ground truth was combined targeted and 12-core systematic transrectal ultrasound-guided biopsy. Four radiologists, two experienced and two less-experienced, evaluated each case twice, once without (DL-CAD-) and once assisted by DL-CAD (DL-CAD+). ROC analysis, sensitivities, specificities, PPV and NPV were calculated to compare the diagnostic accuracy for the diagnosis of prostate cancer (PCa) between the two groups (DL-CAD- vs. DL-CAD+). Spearman's correlation coefficients were evaluated to assess the relationship between PI-RADS category and Gleason score (GS). Also, the median reading times were compared for the two reading groups. In total, 172 patients were included in the final analysis. With DL-CAD assistance, the overall AUC of the less-experienced radiologists increased significantly from 0.66 to 0.80 (p = 0.001; cutoff ISUP GG≥1) and from 0.68 to 0.80 (p = 0.002; cutoff ISUP GG ≥ 2). Experienced radiologists showed an AUC increase from 0.81 to 0.86 (p = 0.146; cutoff ISUP GG ≥ 1) and from 0.81 to 0.84 (p = 0.433; cutoff ISUP GG ≥ 2). Furthermore, the correlation between PI-RADS category and GS improved significantly in the DL-CAD + group (0.45 vs. 0.57; p = 0.03), while the median reading time was reduced from 157 to 150 s (p = 0.023). DL-CAD assistance increased the mean detection performance, with the most significant benefit for the less-experienced radiologist; with the help of DL-CAD less-experienced radiologists reached performances comparable to that of experienced radiologists. • DL-CAD used as a concurrent reading aid helps radiologists to distinguish between benign and cancerous lesions in prostate MRI. • With the help of DL-CAD, less-experienced radiologists may achieve detection performances comparable to that of experienced radiologists. • DL-CAD assistance increases the correlation between PI-RADS category and cancer grade.

Digital Breast Tomosynthesis and Digital Mammography Recall and False-Positive Rates by Time of Day and Reader Experience.

Background Digital breast tomosynthesis (DBT) image interpretation might be more cognitively demanding than interpretation of digital mammography (DM) images. The time of day of interpretation might affect recall and false-positive (FP) rates, especially for DBT. Purpose To determine whether recall and FP rates vary by time of day of interpretation for screening mammography for breast cancer performed with DM and DBT. Materials and Methods This is a retrospective study examining 97 671 screening mammograms interpreted by 18 radiologists between January 2018 and December 2019 at one of 12 community radiology sites. The association between the time of day of interpretation, the type of image interpreted (DM vs DBT), and radiologist experience (≤5 posttraining years vs >5 posttraining years) and the likelihood of a patient being recalled from screening mammography and the likelihood of whether the interpretation was FP or true positive were analyzed. Analyses were conducted using generalized linear mixed modeling with a binary distribution and sandwich estimation where observations were nested by radiologist. Results Screening mammograms interpreted by 18 radiologists were reviewed (40 220 DBTs, 57 451 DMs). Nine radiologists had 5 or fewer posttraining years of experience, and nine had more than 5 posttraining years of experience. The overall recall rates for DM (10.2%) and DBT (9.0%) were different (P = .006); FP rate also differed (9.8% DM, 8.6% DBT; P = .004). For radiologists with 5 or fewer posttraining years of experience, odds of recall increased 11.5% (odds ratio [OR] = 1.12, P = .01) with every hour when using DBT, but this was not found for DM (OR = 1.09, P = .06); DBT and DM were different (OR = 1.12 vs 1.09, P = .02). For radiologists with more than 5 posttraining years of experience, no evidence of increase in recall was observed for DBT (OR = 1.02, P = .27) or DM (OR = 1.0, P = .80), and there was no evidence that these were different (OR = 1.02 vs 1.0, P = .13). Conclusion Patients were more likely to be recalled when their screening digital breast tomosynthesis images were interpreted later in the day by less-experienced radiologists. © RSNA, 2022 Online supplemental material is available for this article.

Variations in breast cancer detection rates during mammogram-reading sessions: does experience have an impact?

To examine whether radiologists' performances are consistent throughout a reading session and whether any changes in performance over the reading task differ depending on experience of the reader. The performance of ten radiologists reading a test set of 60 mammographic cases without breaks was assessed using an ANOVA, 2 × 3 factorial design. Participants were categorized as more (≥2,000 mammogram readings per year) or less (<2,000 readings per year) experienced. Three series of 20 cases were chosen to ensure comparable difficulty and presented in the same sequence to all readers. It usually takes around 30 min for a radiologist to complete each of the 20-case series, resulting in a total of 90 min for the 60 mammographic cases. The sensitivity, specificity, lesion sensitivity, and area under the ROC curve were calculated for each series. We hypothesized that the order in which a series was read (i.e. fixed-series sequence) would have a significant main effect on the participants' performance. We also determined if significant interactions exist between the fixed-series sequence and radiologist experience. Significant linear interactions were found between experience and the fixed sequence of the series for sensitivity (F[1] =5.762, p = .04, partial η2 = .41) and lesion sensitivity. (F[1] =6.993, p = .03, partial η2 = .46). The two groups' mean scores were similar for the first series but progressively diverged. By the end of the third series, significant differences in sensitivity and lesion sensitivity were evident, with the more experienced individuals demonstrating improving and the less experienced declining performance. Neither experience nor series sequence significantly affected the specificity or the area under the ROC curve. Radiologists' performance may change considerably during a reading session, apparently as a function of experience, with less experienced radiologists declining in sensitivity and lesion sensitivity while more experienced radiologists actually improve. With the increasing demands on radiologists to undertake high-volume reporting, we suggest that junior radiologists be made aware of possible sensitivity and lesion sensitivity deterioration over time so they can schedule breaks during continuous reading sessions that are appropriate to them, rather than try to emulate their more experienced colleagues. Less-experienced radiologists demonstrated a reduction in mammographic diagnostic accuracy in later stages of the reporting sessions. This may suggest that extending the duration of reporting sessions to compensate for increasing workloads may not represent the optimal solution for less-experienced radiologists.

Detection and PI-RADS classification of focal lesions in prostate MRI: Performance comparison between a deep learning-based algorithm (DLA) and radiologists with various levels of experience

PurposeTo compare the performance of lesion detection and Prostate Imaging-Reporting and Data System (PI-RADS) classification between a deep learning-based algorithm (DLA), clinical reports and radiologists with different levels of experience in prostate MRI. MethodsThis retrospective study included 121 patients who underwent prebiopsy MRI and prostate biopsy. More than five radiologists (Reader groups 1, 2: residents; Readers 3, 4: less-experienced radiologists; Reader 5: expert) independently reviewed biparametric MRI (bpMRI). The DLA results were obtained using bpMRI. The reference standard was based on pathologic reports. The diagnostic performance of the PI-RADS classification of DLA, clinical reports, and radiologists was analyzed using AUROC. Dichotomous analysis (PI-RADS cutoff value ≥ 3 or 4) was performed, and the sensitivities and specificities were compared using McNemar’s test. ResultsClinically significant cancer [CSC, Gleason score ≥ 7] was confirmed in 43 patients (35.5%). The AUROC of the DLA (0.828) for diagnosing CSC was significantly higher than that of Reader 1 (AUROC, 0.706; p = 0.011), significantly lower than that of Reader 5 (AUROC, 0.914; p = 0.013), and similar to clinical reports and other readers (p = 0.060–0.661). The sensitivity of DLA (76.7%) was comparable to those of all readers and the clinical reports at a PI-RADS cutoff value ≥ 4. The specificity of the DLA (85.9%) was significantly higher than those of clinical reports and Readers 2–3 and comparable to all others at a PI-RADS cutoff value ≥ 4. ConclusionsThe DLA showed moderate diagnostic performance at a level between those of residents and an expert in detecting and classifying according to PI-RADS. The performance of DLA was similar to that of clinical reports from various radiologists in clinical practice.

Inter-reader agreement of the Society of Abdominal Radiology-American Gastroenterological Association (SAR-AGA) consensus reporting for key phenotypes at MR enterography in adults with Crohn disease: impact of radiologist experience.

To assess inter-reader agreement of key features from the SAR-AGA recommendations for the interpretation and reporting of MRE in adult patients with CD, focusing on the impact of radiologist experience on inter-reader agreement of CD phenotypes. Two experienced and two less-experienced radiologists retrospectively evaluated 99 MRE in CD patients (50 initial MRE, 49 follow-up MRE) performed from 1/1/2019 to 3/20/2020 for the presence of active bowel inflammation (stomach, proximal small bowel, ileum, colon), stricture, probable stricture, penetrating disease, and perianal disease. The MRE protocol did not include dedicated perianal sequences. Inter-rater agreement was determined for each imaging feature using prevalence-adjusted bias-adjusted kappa and compared by experience level. All readers had almost-perfect inter-reader agreement (κ > 0.90) for penetrating disease, abscess, and perianal abscess in all 99 CD patients. All readers had strong inter-reader agreement (κ: 0.80-0.90) in 99 CD patients for active ileum inflammation, proximal small bowel inflammation, and stricture. Less-experienced readers had significantly lower inter-reader agreement for active ileum inflammation on initial than follow-up MRE (κ 0.68 versus 0.96, p = 0.018) and for strictures on follow-up than initial MRE (κ 0.76 versus 1.0, p = 0.027). Experienced readers had significantly lower agreement for perianal fistula on follow-up than initial MRE (κ: 0.55 versus 0.92, p = 0.008). There was strong to almost-perfect inter-reader agreement for key CD phenotypes described in the SAR-AGA consensus recommendations including active ileum and proximal small bowel inflammation, stricture, penetrating disease, abscess, and perianal abscess. Areas of lower inter-reader agreement could be targeted for future education efforts to further standardize CD MRE reporting. Dedicated perianal sequences should be included on follow-up MRE.

Multiparametric transvaginal ultrasound in the diagnosis of endometrial cancer in post-menopausal bleeding: diagnostic performance of a transvaginal algorithm and reproducibility amongst less experienced observers.

(a) To comparatively evaluate the performance of grayscale ultrasound features, power Doppler (PD) blood flow characteristics, and gel infusion sonography (GIS) in diagnosing endometrial cancer during real-time examination, (b) to compare the performance of real-time diagnosis of endometrial cancer by experienced observers with offline analysis by blinded observers using similar sonographic criteria during review of cine loop clips. 152 females with post-menopausal bleeding (PMB) had ET ≥ 4 mm at first-line ultrasound were included. Two experienced radiologists evaluated endometrial patterns at real-time evaluation (grayscale ultrasound, PD, and GIS), then examinations were stored as video clips for later evaluation by two less-experienced radiologists. The reference standard was hysteroscopy (HY) and/or hysterectomy with the histopathological examination. The area under (AUC) the receiver operating characteristic (ROC) curve was calculated to assess the diagnostic performance for the prediction of endometrial cancer. Among 152 females with ET ≥ 4 mm at first line TVUS, 88 (57.9%) patients had endometrial cancer on final pathologic analysis. Real-time ultrasound criteria (ET ≥ 5 mm with the presence of irregular branching endometrial blood vessels or multiple vessels crossing EM or areas with densely packed color-splash vessels with non-intact or interrupted EMJ at the grayscale ultrasound and/or GIS) correctly diagnosed 95% of endometrial cancers with 92% diagnostic efficiency.There is comparable accuracy of real-time evaluation (96%) and offline analysis (92%) after the exclusion of poor quality videos from the analysis. The diagnostic criteria showed good to an excellent agreement between real-time ultrasound and offline analysis. When real-time ultrasound is performed with good technique, utilizing multiple parameters, it is possible to diagnose endometrial cancer with a high degree of accuracy and reproducibility. when real-time ultrasound is performed with good technique, utilizing multiple parameters, it is possible to diagnose endometrial cancer with a high degree of accuracy and reproducibility.

Diagnostic performance for pulmonary adenocarcinoma on CT: comparison of radiologists with and without three-dimensional convolutional neural network.

To compare diagnostic performance for pulmonary invasive adenocarcinoma among radiologists with and without three-dimensional convolutional neural network (3D-CNN). Enrolled were 285 patients with adenocarcinoma in situ (AIS, n = 75), minimally invasive adenocarcinoma (MIA, n = 58), and invasive adenocarcinoma (IVA, n = 152). A 3D-CNN model was constructed with seven convolution-pooling and two max-pooling layers and fully connected layers, in which batch normalization, residual connection, and global average pooling were used. Only the flipping process was performed for augmentation. The output layer comprised two nodes for two conditions (AIS/MIA and IVA) according to prognosis. Diagnostic performance of the 3D-CNN model in 285 patients was calculated using nested 10-fold cross-validation. In 90 of 285 patients, results from each radiologist (R1, R2, and R3; with 9, 14, and 26 years of experience, respectively) with and without the 3D-CNN model were statistically compared. Without the 3D-CNN model, accuracy, sensitivity, and specificity of the radiologists were as follows: R1, 70.0%, 52.1%, and 90.5%; R2, 72.2%, 75%, and 69%; and R3, 74.4%, 89.6%, and 57.1%, respectively. With the 3D-CNN model, accuracy, sensitivity, and specificity of the radiologists were as follows: R1, 72.2%, 77.1%, and 66.7%; R2, 74.4%, 85.4%, and 61.9%; and R3, 74.4%, 93.8%, and 52.4%, respectively. Diagnostic performance of each radiologist with and without the 3D-CNN model had no significant difference (p > 0.88), but the accuracy of R1 and R2 was significantly higher with than without the 3D-CNN model (p < 0.01). The 3D-CNN model can support a less-experienced radiologist to improve diagnostic accuracy for pulmonary invasive adenocarcinoma without deteriorating any diagnostic performances. • The 3D-CNN model is a non-invasive method for predicting pulmonary invasive adenocarcinoma in CT images with high sensitivity. • Diagnostic accuracy by a less-experienced radiologist was better with the 3D-CNN model than without the model.

Incidental findings on emergency CT scans: Predictive factors and medico-economic impact

PurposeThe main objective was to evaluate types and predictive factors of incidental findings (IFs) on multidetector computed tomographies (MDCTs) performed for an emergency department (ED). The secondary aim was to analyze additional investigations, their benefits, side effects, costs and the final diagnoses. MethodOne thousand consecutive patients over 18 years old who underwent an MDCT in the ED of our institution from January 2011 to November 2011 were retrospectively included, accounting for 300 head MDCTs and 700 other MDCTs. The following criteria were collected in patient electronic medical records: IFs (divided into low and high clinical significance), body areas covered, availability of a prior imaging, radiologist’s experience and subspecialty, additional investigations, their outcomes and costs. ResultsAmong the 1000 included patients, 232 had at least one IF and 122 had at least one IF of high clinical significance (IFCS). There were 340 IFs and 150 IFCSs. A significant association with the presence of at least one IF was noted for older patients, less-experienced radiologists, no subspecialty of the radiologist, the abdominopelvic area, and the absence of prior imaging. Eighteen IFs generated additional investigations in our institution, including five invasive samplings and three surgical operations, with two diagnoses of malignancy (a gastrointestinal stromal tumor and a Bosniak IV cystic renal lesion). One benign iatrogenic complication occurred. Total cost of these investigations was €41,247 (with an average of €2292 per IF investigated). ConclusionIFs on emergency MDCTs were frequent, rarely severe, rarely iatrogenic and relatively expensive.

Improving B-Mode Ultrasound &amp;nbsp;Diagnostic Performance&amp;nbsp;for Focal Liver Lesions Using Deep Learning:&amp;nbsp;A Prospective&amp;nbsp;Multicentre&amp;nbsp;Study

Background: The diagnosis performance of B-mode ultrasound (US) for focal liver lesions (FLLs) is relatively limited. The aim of this study was to develop a deep convolutional neural network of US (DCNN-US) for aiding radiologists in classification of malignant from benign FLLs with high sensitivity and specificity. Methods: This prospective multicentre study was conducted in 13 hospitals with different levels in China. Patients who had non-cystic FLLs with pathological results were enrolled. The FLLs from 11 hospitals were randomly divided into training and internal validations (IV) cohorts with a 4:1 ratio for developing and evaluating DCNN-US. Diagnostic performance of the model was verified using external validation (EV) cohort from another two hospitals. Three models were developed by using lesion (ModelLesion, ModelL), liver background images (Model Lesion+Background, ModelL B) and clinical factors (ModelLesion+Background +clinic, ModelL BC) to diagnose FLLs, respectively. The diagnosis value of DCNN-US was compared with that of contrast enhanced computed tomography (CT)/magnetic resonance image (MRI) and 236 radiologists, respectively. Findings: We enrolled 24,343 US images of 2,143 patients to develop and test DCNN-US model. The diagnostic area under the curve (AUC) values of ModelL B increased 6% than ModelL, and Model L BC increased nearly 11% than ModelL B. The AUC of ModelLBC for FLLs were 0.925 (95% CI: 0.886–0.963), and 0.924 (95% CI: 0.889–0.959) in the IV and EV cohorts, respectively. The diagnostic sensitivity and specificity of ModelL BC were superior to 15-year skilled radiologists (86.5% vs 76.1%, and 85.5% vs 75.9%, respectively) (both p<0.01). Accuracy of ModelL BC was comparable to that of contrast enhanced CT (both 84.7%) but inferior to contrast enhanced MRI (87.9%). Interpretation: DCNN-US with high sensitivity and specificity in diagnosing FLLs shows its potential to assist less-experienced radiologists in improving their performance and lowering their dependence on sectional imaging in liver cancer diagnosis. Trial Registration: ClinicalTrials.gov (NCT03871140). Funding Statement: The National Social Science Foundation, the National Key R&D Program, the National Scientific Foundation Committee of Beijing, Ministry of Science and Technology, the Strategic Priority Research Program of China, Chinese Academy of Sciences and Beijing Municipal Science & Technology Commission. Declaration of Interests: No conflicts of interest. Ethics Approval Statement: This prospective multicentre study was approved by the ethics committee of each centre.

Computer-Aided Diagnosis System for the Evaluation of Thyroid Nodules on Ultrasonography: Prospective Non-Inferiority Study according to the Experience Level of Radiologists.

ObjectiveTo determine whether a computer-aided diagnosis (CAD) system for the evaluation of thyroid nodules is non-inferior to radiologists with different levels of experience.Materials and MethodsPatients with thyroid nodules with a decisive diagnosis of benign or malignant nodule were consecutively enrolled from November 2017 to September 2018. Three radiologists with different levels of experience (1 month, 4 years, and 7 years) in thyroid ultrasound (US) reviewed the thyroid US with and without using the CAD system. Statistical analyses included non-inferiority testing of the diagnostic accuracy for malignant thyroid nodules between the CAD system and the three radiologists with a non-inferiority margin of 10%, comparison of the diagnostic performance, and the added value of the CAD system to the radiologists.ResultsAltogether, 197 patients were included in the study cohort. The diagnostic accuracy of the CAD system (88.48%, 95% confidence interval [CI] = 82.65–92.53) was non-inferior to that of the radiologists with less experience (1 month and 4 year) of thyroid US (83.03%, 95% CI = 76.52–88.02; p < 0.001), whereas it was inferior to that of the experienced radiologist (7 years) (95.76%, 95% CI = 91.37–97.96; p = 0.138). The sensitivity and negative predictive value of the CAD system were significantly higher than those of the less-experienced radiologists were, whereas no significant difference was found with those of the experienced radiologist. A combination of US and the CAD system significantly improved sensitivity and negative predictive value, although the specificity and positive predictive value deteriorated for the less-experienced radiologists.ConclusionThe CAD system may offer support for decision-making in the diagnosis of malignant thyroid nodules for operators who have less experience with thyroid US.

Can eye-tracking metrics be used to better pair radiologists in a mammogram reading task?

To propose a framework for optimal pairing of radiologists when reading mammograms based on their search patterns. Four experienced and four less-experienced radiologists were asked to assess 120 cases (59 with cancers) while their eye positions were tracked. Fourteen eye-tracking metrics were extracted to quantify the differences among radiologists' visual search pattern. For each radiologist and metric, less-experienced radiologists and expert readers were ranked based on the level of similarities in gaze patterns (from the most different to the most similar). Less-experienced readers and experts were also ranked based on the values of area under the receiver operating characteristic curve (AUC) after pairing (the best possible way of ranking). Using the Kendall's tau distance, rankings based on different metrics were compared with the best possible ranking. Using paired Wilcoxon signed-rank test, the AUC values when pairing in the best way were compared with pairing based on different metrics. Finally, we investigated the robustness of pairing strategies against the small sample size. For ranking the experienced radiologists, results from eight metrics were as good as the best possible ranking. For the less-experienced ones, only one metric resulted in a ranking comparable to the best possible way of ranking. The AUC values of pairings based on these metrics did not differ significantly from the best pairing scenario. Compared to the pairings based on the cognitive metrics, the ranking based on AUC values varied more greatly with the sample size, suggesting that it is less robust against the small sample size compared to the cognitive metrics. Different pairings may have different effects on performance; some are detrimental while some improve the performance of the pair. Using the suggested cognitive metrics, we can optimize the pairings even with a small dataset.

Characterization of Prostate Cancer with Gleason Score of at Least 7 by Using Quantitative Multiparametric MR Imaging: Validation of a Computer-aided Diagnosis System in Patients Referred for Prostate Biopsy.

Purpose To determine the performance of a computer-aided diagnosis (CAD) system trained at characterizing cancers in the peripheral zone (PZ) with a Gleason score of at least 7 in patients referred for multiparametric magnetic resonance (MR) imaging before prostate biopsy. Materials and Methods Two institutional review board-approved prospective databases of patients who underwent multiparametric MR imaging before prostatectomy (database 1) or systematic and targeted biopsy (database 2) were retrospectively used. All patients gave informed consent for inclusion in the databases. A CAD combining the 10th percentile of the apparent diffusion coefficient and the time to peak of enhancement was trained to detect cancers in the PZ with a Gleason score of at least 7 in 106 patients from database 1. The CAD was tested in 129 different patients from database 2. All targeted lesions were prospectively scored at biopsy by using a five-level Likert score. The CAD scores were retrospectively calculated. Biopsy results were used as the reference standard. Areas under the receiver operating characteristic curves (AUCs) were computed for CAD and Likert scores by using binormal smoothing for per-lesion and per-lobe analyses, and a density function for per-patient analysis. Results The CAD outperformed the Likert score in the overall population and all subgroups, except in the transition zone. The difference was statistically significant for the overall population (AUC, 0.95 [95% confidence interval {CI}: 0.90, 0.98] vs 0.88 [95% CI: 0.68, 0.96]; P = .02) at per-patient analysis, and for less-experienced radiologists (<1 year) at per-lesion (AUC, 0.90 [95% CI: 0.81, 0.95] vs 0.83 [95% CI: 0.73, 0.90]; P = .04) and per-lobe (AUC, 0.92 [95% CI: 0.80, 0.96] vs 0.84 [95% CI: 0.72, 0.91]; P = .04) analysis. Conclusion The CAD outperformed the Likert score prospectively assigned at biopsy in characterizing cancers with a Gleason score of at least 7. © RSNA, 2018 Online supplemental material is available for this article.

Disproportionate fluid sign as an aid in diagnosing high-grade bursal-sided supraspinatus tendon tear

Background Deep, high-grade bursal-sided supraspinatus tendon tears are sometimes preoperatively misinterpreted as full-thickness tears on shoulder magnetic resonance imaging (MRI). Purpose To determine the usefulness of disproportionate fluid sign for differentiating high-grade bursal-sided partial-thickness tears from full-thickness tears on conventional MRI. Material and Methods Preoperative MRIs of 198 patients with arthroscopically confirmed high-grade bursal-sided partial-thickness tears and full-thickness tears were independently reviewed by two readers on two occasions. The presence of high-grade bursal-sided partial-thickness tears with a confidence level using a five-point grading scale was assessed based on tear depth alone and also in combination with disproportionate fluid sign, defined as a prominent subdeltoid or subacromial-subdeltoid bursal fluid distension with a relative paucity of effusion in the glenohumeral joint. The sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC) were calculated, as well as inter-observer reliability. Results The disproportionate fluid sign was identified in 60/74 (81.2%) bursal-sided partial-thickness tears and 9/124 (7.5%) full-thickness tears. The sensitivity and accuracy of the diagnosis of bursal-sided tear were higher when disproportionate fluid sign was used in conjunction with the tear depth, compared with tear depth alone ( P < 0.001). There was excellent inter-observer agreement for disproportionate fluid sign and deep bursal-sided tear. The AUCs were significantly higher in combination with disproportionate fluid sign. Conclusion The disproportionate fluid sign indicates the presence of a deep, high-grade bursal-sided partial-thickness tear, which can be misinterpreted as a full-thickness tear. Thus, it can provide greater diagnostic assistance to less-experienced radiologists and clinicians.

Additive value of "otosclerosis-weighted" images for the CT diagnosis of fenestral otosclerosis.

Background Otosclerotic foci are usually seen as minute low-density lesions and this may be attributed to relatively low sensitivity on visual assessment using computed tomography (CT). Otosclerotic foci can be detected by using the accurate region of interest (ROI) setting, while small ROI settings by less-experienced radiologists may result in false negative findings. Purpose To evaluate the diagnostic ability of our proposed method ("otosclerosis-weighted" imaging [OWI]), which is based on reversing the density, compared with conventional CT (CCT) imaging alone. Material and Methods Temporal bone CTs of consecutive patients with otosclerosis were analyzed. Gender- and age-matched control participants were also included. All CT images were obtained using a 64-detector row scanner. OWI was obtained by extracting the temporal bone region using the threshold technique and reversing the density (black to white). Four independent radiologists took part in two reading sessions. In the first session, the observers read only CCT imaging. In the second session, they read OWI along with the CCT imaging. Sensitivity was assessed for the four readers. Results Thirty temporal bones of 25 patients with otosclerosis (3 men, 22 women; mean age, 53.9 ± 9.0 years) and 30 temporal bones of 30 control participants (4 men, 26 women; mean age, 44.0 ± 16.2 years) were included. For all observers, reading with a combination of the two methods was associated with a higher sensitivity (63.3-80.0%) than with conventional CT images alone (30.0-60.0%; P < 0.05, each). Conclusion Application of our proposed method based on threshold value may help detect foci of fenestral otosclerosis.