Junior Radiologists Research Articles

Assessing the inflammatory severity of the terminal ileum in Crohn disease using radiomics based on MRI

BackgroundEvaluating inflammatory severity using imaging is essential for Crohn’s disease, but it is limited by potential interobserver variation and subjectivity. We compared the efficiency of magnetic resonance index of activity (MaRIA) collected by radiologists and a radiomics model in assessing the inflammatory severity of terminal ileum (TI).Methods121 patients were collected from two centers. Patients were divided into ulcerative group and mucosal remission group based on the TI Crohn’s disease Endoscopic Severity Index. The consistency of bowel wall thickness (BWT), relative contrast enhancement (RCE), edema, ulcer, MaRIA and features of the region of interest between radiologists were described by weighted Kappa test and intraclass correlation coefficient (ICC), and developed receiver operating curve of MaRIA. The radiomics model was established using reproducible features of logistic regression based on arterial staging of T1WI sequences. Delong test was used to compare radiomics with MaRIA.ResultsThe consistency between radiologists were moderate in BWT (ICC = 0.638), fair in edema (κ = 0.541), RCE (ICC = 0.461), MaRIA (ICC = 0.579) and poor in ulcer (κ = 0.271). Radiomics model was developed by 6 reproducible features (ICC = 0.93–0.96) and equivalent to MaRIA which evaluated by the senior radiologist (0.872 vs 0.883 in training group, 0.824 vs 0.783 in validation group, P = 0.847, 0.471), both of which were significantly higher than MaRIA evaluated by junior radiologist (AUC: 0.621 in training group, 0.557 in validation group, all, P < 0.05).ConclusionThe evaluation of inflammatory severity could be performed by radiomics objectively and reproducibly, and was comparable to MaRIA evaluated by the senior radiologist. Radiomics may be an important method to assist junior radiologists to assess the severity of inflammation objectively and accurately.

Lower-extremity fatigue fracture detection and grading based on deep learning models of radiographs.

To identify the feasibility of deep learning-based diagnostic models for detecting and assessing lower-extremity fatigue fracture severity on plain radiographs. This retrospective study enrolled 1151 X-ray images (tibiofibula/foot: 682/469) of fatigue fractures and 2842 X-ray images (tibiofibula/foot: 2000/842) without abnormal presentations from two clinical centers. After labeling the lesions, images in a center (tibiofibula/foot: 2539/1180) were allocated at 7:1:2 for model construction, and the remaining images from another center (tibiofibula/foot: 143/131) for external validation. A ResNet-50 and a triplet branch network were adopted to construct diagnostic models for detecting and grading. The performances of detection models were evaluated with sensitivity, specificity, and area under the receiver operating characteristic curve (AUC), while grading models were evaluated with accuracy by confusion matrix. Visual estimations by radiologists were performed for comparisons with models. For the detection model on tibiofibula, a sensitivity of 95.4%/85.5%, a specificity of 80.1%/77.0%, and an AUC of 0.965/0.877 were achieved in the internal testing/external validation set. The detection model on foot reached a sensitivity of 96.4%/90.8%, a specificity of 76.0%/66.7%, and an AUC of 0.947/0.911. The detection models showed superior performance to the junior radiologist, comparable to the intermediate or senior radiologist. The overall accuracy of the diagnostic model was 78.5%/62.9% for tibiofibula and 74.7%/61.1% for foot in the internal testing/external validation set. The deep learning-based models could be applied to the radiological diagnosis of plain radiographs for assisting in the detection and grading of fatigue fractures on tibiofibula and foot. • Fatigue fractures on radiographs are relatively difficult to detect, and apt to be misdiagnosed. • Detection and grading models based on deep learning were constructed on a large cohort of radiographs with lower-extremity fatigue fractures. • The detection model with high sensitivity would help to reduce the misdiagnosis of lower-extremity fatigue fractures.

Combined artificial intelligence and radiologist model for predicting rectal cancer treatment response from magnetic resonance imaging: an external validation study.

To evaluate an MRI-based radiomic texture classifier alone and combined with radiologist qualitative assessment in predicting pathological complete response (pCR) using restaging MRI with internal training and external validation. Consecutive patients with locally advanced rectal cancer (LARC) who underwent neoadjuvant therapy followed by total mesorectal excision from March 2012 to February 2016 (Memorial Sloan Kettering Cancer Center/internal dataset, n = 114, 41% female, median age = 55) and July 2014 to October 2015 (Instituto do Câncer do Estado de São Paulo/external dataset, n = 50, 52% female, median age = 64.5) were retrospectively included. Two radiologists (R1, senior; R2, junior) independently evaluated restaging MRI, classifying patients (radiological complete response vs radiological partial response). Model A (n = 33 texture features), model B (n = 91 features including texture, shape, and edge features), and two combination models (model A + B + R1, model A + B + R2) were constructed. Pathology served as the reference standard for neoadjuvant treatment response. Comparison of the classifiers' AUCs on the external set was done using DeLong's test. Models A and B had similar discriminative ability (P = 0.3; Model B AUC = 83%, 95% CI 70%-97%). Combined models increased inter-reader agreement compared with radiologist-only interpretation (κ = 0.82, 95% CI 0.70-0.89 vs k = 0.25, 95% CI 0.11-0.61). The combined model slightly increased junior radiologist specificity, positive predictive value, and negative predictive values (93% vs 90%, 57% vs 50%, and 91% vs 90%, respectively). We developed and externally validated a combined model using radiomics and radiologist qualitative assessment, which improved inter-reader agreement and slightly increased the diagnostic performance of the junior radiologist in predicting pCR after neoadjuvant treatment in patients with LARC.

How to improve the performance of acetabulum fracture diagnoses: investigating the use of three-dimensional CT-scan reconstructions.

The classification of acetabular fractures remains a challenge for the junior radiologist, although he is the first line of diagnosis in Emergency Department. The advantages of three-dimensional reconstructions have yet to be evaluated on a large scale. A total of 212 fractures were classified according to Letournel and Judet by a senior orthopaedic surgeon, a senior radiologist, and a resident radiologist. The CT scans were first analysed in 2D and then a second time using 2D + 3D reconstructions of the acetabulum excluding the femoral head. 3D reconstructions improved correct classification not only for the radiologist resident (+ 5%) but also for senior radiologist and orthopaedist (+ 2 and + 3%). 3D reconstructions also more significantly improved the diagnoses of complex fractures (+ 8.3%) compared to simple fractures (+ 0.4%). 3D reconstructions have improved the routine diagnosis of acetabular fractures, especially for junior radiologist or in cases of complex fractures.

Space-Occupying Lesions of the Inner Ear Are Easily Misdiagnosed as Endolymphatic Hydrops in a Perilymph-Enhanced Sequence Without the Assistance of a Heavily T2-Weighted Sequence.

The aim of the study was to explore the value of T2-sampling perfection with application-optimized contrasts by using different flip angle evolutions (T2-SPACE) in identifying space-occupying lesions of the inner ear. We collected the T2-SPACE and 3-dimensional inversion-recovery sequence with real reconstruction (3D-real IR) images of 220 patients with inner ear symptoms, including 15 patients with inner ear space-occupying lesions. With T2-SPACE images hidden, a senior and junior radiologist made a diagnosis for all patients using only the 3D-real IR images. After 4 weeks the images were shuffled, and T2-SPACE images were made available to the 2 radiologists in addition to 3D-real IR to reconsider the diagnosis for all patients. With the SPACE images hidden, the correct diagnosis rate of the space-occupying lesions was 8/15 (53.3%) for the senior radiologist, whereas it was only 2/15 (13.3%) for the junior radiologist. Without the SPACE images hidden, the correct diagnosis rate of the space-occupying lesions was 15/15 (100.0%) for the senior radiologist, whereas it was 13/15 (86.7%) for the junior radiologist. Of the 15 patients, 7 had only vestibular space-occupying lesions, 2 had only cochlear space-occupying lesions, and 6 had both. No semicircular canal space-occupying lesion was observed. T2-SPACE can help identify space-occupying lesions of the inner ear that tend to be misdiagnosed as endolymphatic hydrops on 3D-real IR. The senior radiologist had a higher rate for the identification of space-occupying lesions than the junior radiologist when using only 3D-real IR, although the senior radiologist detection rate was still only 53.3%. With the addition of T2-SPACE, both the junior and senior radiologist achieved a high detection rate, which increased to 86.7% and 100%, respectively.

Diagnostic study on clinical feasibility of an AI-based diagnostic system as a second reader on mobile CT images: a preliminary result.

BackgroundArtificial intelligence (AI) has breathed new life into the lung nodules detection and diagnosis. However, whether the output information from AI will translate into benefits for clinical workflow or patient outcomes in a real-world setting remains unknown. This study was to demonstrate the feasibility of an AI-based diagnostic system deployed as a second reader in imaging interpretation for patients screened for pulmonary abnormalities in a clinical setting.MethodsThe study included patients from a lung cancer screening program conducted in Sichuan Province, China using a mobile computed tomography (CT) scanner which traveled to medium-size cities between July 10th, 2020 and September 10th, 2020. Cases that were suspected to have malignant nodules by junior radiologists, senior radiologists or AI were labeled a high risk (HR) tag as HR-junior, HR-senior and HR-AI, respectively, and included into final analysis. The diagnosis efficacy of the AI was evaluated by calculating negative predictive value and positive predictive value when referring to the senior readers’ final results as the gold standard. Besides, characteristics of the lesions were compared among cases with different HR labels.ResultsIn total, 251/3,872 patients (6.48%, male/female: 91/160, median age, 66 years) with HR lung nodules were included. The AI algorithm achieved a negative predictive value of 88.2% [95% confidence interval (CI): 62.2–98.0%] and a positive predictive value of 55.6% (95% CI: 49.0–62.0%). The diagnostic duration was significantly reduced when AI was used as a second reader (223±145.6 vs. 270±143.17 s, P<0.001). The information yielded by AI affected the radiologist’s decision-making in 35/145 cases. Lesions of HR cases had a higher volume [309.9 (214.9–732.5) vs. 141.3 (79.3–380.8) mm3, P<0.001], lower average CT number [−511.0 (−576.5 to −100.5) vs. −191.5 (−487.3 to 22.5), P=0.010], and pure ground glass opacity rather than solid.ConclusionsThe AI algorithm had high negative predictive value but low positive predictive value in diagnosing HR lung lesions in a clinical setting. Deploying AI as a second reader could help avoid missed diagnoses, reduce diagnostic duration, and strengthen diagnostic confidence for radiologists.

Automatic detection and voxel-wise mapping of lumbar spine Modic changes with deep learning.

BackgroundModic changes (MCs) are the most prevalent classification system for describing magnetic resonance imaging (MRI) signal intensity changes in the vertebrae. However, there is a growing need for novel quantitative and standardized methods of characterizing these anomalies, particularly for lesions of transitional or mixed nature, due to the lack of conclusive evidence of their associations with low back pain. This retrospective imaging study aims to develop an interpretable deep learning‐based detection tool for voxel‐wise mapping of MCs.MethodsSeventy‐five lumbar spine MRI exams that presented with acute‐to‐chronic low back pain, radiculopathy, and other symptoms of the lumbar spine were enrolled. The pipeline consists of two deep convolutional neural networks to generate an interpretable voxel‐wise Modic map. First, an autoencoder was trained to segment vertebral bodies from T1‐weighted sagittal lumbar spine images. Next, two radiologists segmented and labeled MCs from a combined T1‐ and T2‐weighted assessment to serve as ground truth for training a second autoencoder that performs segmentation of MCs. The voxels in the detected regions were then categorized to the appropriate Modic type using a rule‐based signal intensity algorithm. Post hoc, three radiologists independently graded a second dataset with the aid of the model predictions in an artificial (AI)‐assisted experiment.ResultsThe model successfully identified the presence of changes in 85.7% of samples in the unseen test set with a sensitivity of 0.71 (±0.072), specificity of 0.95 (±0.022), and Cohen's kappa score of 0.63. In the AI‐assisted experiment, the agreement between the junior radiologist and the senior neuroradiologist significantly improved from Cohen's kappa score of 0.52 to 0.58 (p < 0.05).ConclusionsThis deep learning‐based approach demonstrates substantial agreement with radiologists and may serve as a tool to improve inter‐rater reliability in the assessment of MCs.

MRI-Based Radiomics Models to Discriminate Hepatocellular Carcinoma and Non-Hepatocellular Carcinoma in LR-M According to LI-RADS Version 2018.

Differentiating hepatocellular carcinoma (HCC) from other primary liver malignancies in the Liver Imaging Reporting and Data System (LI-RADS) M (LR-M) tumours noninvasively is critical for patient treatment options, but visual evaluation based on medical images is a very challenging task. This study aimed to evaluate whether magnetic resonance imaging (MRI) models based on radiomics features could further improve the ability to classify LR-M tumour subtypes. A total of 102 liver tumours were defined as LR-M by two radiologists based on LI-RADS and were confirmed to be HCC (n = 31) and non-HCC (n = 71) by surgery. A radiomics signature was constructed based on reproducible features using the max-relevance and min-redundancy (mRMR) and least absolute shrinkage and selection operator (LASSO) logistic regression algorithms with tenfold cross-validation. Logistic regression modelling was applied to establish different models based on T2-weighted imaging (T2WI), arterial phase (AP), portal vein phase (PVP), and combined models. These models were verified independently in the validation cohort. The area under the curve (AUC) of the models based on T2WI, AP, PVP, T2WI + AP, T2WI + PVP, AP + PVP, and T2WI + AP + PVP were 0.768, 0.838, 0.778, 0.880, 0.818, 0.832, and 0.884, respectively. The combined model based on T2WI + AP + PVP showed the best performance in the training cohort and validation cohort. The discrimination efficiency of each radiomics model was significantly better than that of junior radiologists’ visual assessment (p < 0.05; Delong). Therefore, the MRI-based radiomics models had a good ability to discriminate between HCC and non-HCC in LR-M tumours, providing more options to improve the accuracy of LI-RADS classification.

Multiphase Dual-Energy Spectral CT-Based Deep Learning Method for the Noninvasive Prediction of Head and Neck Lymph Nodes Metastasis in Patients With Papillary Thyroid Cancer.

PurposeTo develop deep learning (DL) models based on multiphase dual-energy spectral CT for predicting lymph nodes metastasis preoperatively and noninvasively in papillary thyroid cancer patients.MethodsA total of 293 lymph nodes from 78 papillary thyroid cancer patients who underwent dual-energy spectral CT before lymphadenectomy were enrolled in this retrospective study. The lymph nodes were randomly divided into a development set and an independent testing set following a 4:1 ratio. Four single-modality DL models based on CT-A model, CT-V model, Iodine-A model and Iodine-V model and a multichannel DL model incorporating all modalities (Combined model) were proposed for the prediction of lymph nodes metastasis. A CT-feature model was also built on the selected CT image features. The model performance was evaluated with respect to discrimination, calibration and clinical usefulness. In addition, the diagnostic performance of the Combined model was also compared with four radiologists in the independent test set.ResultsThe AUCs of the CT-A, CT-V, Iodine-A, Iodine-V and CT-feature models were 0.865, 0.849, 0.791, 0.785 and 0.746 in the development set and 0.830, 0.822, 0.744, 0.739 and 0.732 in the testing set. The Combined model had outperformed the other models and achieved the best performance with AUCs yielding 0.890 in the development set and 0.865 in the independent testing set. The Combined model showed good calibration, and the decision curve analysis demonstrated that the net benefit of the Combined model was higher than that of the other models across the majority of threshold probabilities. The Combined model also showed noninferior diagnostic capability compared with the senior radiologists and significantly outperformed the junior radiologists, and the interobserver agreement of junior radiologists was also improved after artificial intelligence assistance.ConclusionThe Combined model integrating both CT images and iodine maps of the arterial and venous phases showed good performance in predicting lymph nodes metastasis in papillary thyroid cancer patients, which could facilitate clinical decision-making.

Evaluation of the Combination of Artificial Intelligence and Radiologist Assessments to Interpret Malignant Architectural Distortion on Mammography.

PurposeTo compare the mammographic malignant architectural distortion (AD) detection performance of radiologists who read mammographic examinations unaided versus those who read these examinations with the support of artificial intelligence (AI) systems.Material and MethodsThis retrospective case-control study was based on a double-reading of clinical mammograms between January 2011 and December 2016 at a large tertiary academic medical center. The study included 177 malignant and 90 benign architectural distortion (AD) patients. The model was built based on the ResNeXt-50 network. Algorithms used deep learning convolutional neural networks, feature classifiers, image analysis algorithms to depict AD and output a score that translated to malignant. The accuracy for malignant AD detection was evaluated using area under the curve (AUC).ResultsThe overall AUC was 0.733 (95% CI, 0.673-0.792) for Reader First-1, 0.652 (95% CI, 0.586-0.717) for Reader First-2, and 0.655 (95% CI, 0.590-0.719) for Reader First-3. and the overall AUCs for Reader Second-1, 2, 3 were 0.875 (95% CI, 0.830-0.919), 0.882 (95% CI, 0.839-0.926), 0.884 (95% CI, 0.841-0.927),respectively. The AUCs for all the reader-second radiologists were significantly higher than those for all the reader-first radiologists (Reader First-1 vs. Reader Second-1, P= 0.004). The overall AUC was 0.792 (95% CI, 0.660-0.925) for AI algorithms. The combination assessment of AI algorithms and Reader First-1 achieved an AUC of 0.880 (95% CI, 0.793-0.968), increased than the Reader First-1 alone and AI algorithms alone. AI algorithms alone achieved a specificity of 61.1% and a sensitivity of 80.6%. The specificity for Reader First-1 was 55.5%, and the sensitivity was 86.1%. The results of the combined assessment of AI and Reader First-1 showed a specificity of 72.7% and sensitivity of 91.7%. The performance showed significant improvements compared with AI alone (p<0.001) as well as the reader first-1 alone (p=0.006).ConclusionWhile the single AI algorithm did not outperform radiologists, an ensemble of AI algorithms combined with junior radiologist assessments were found to improve the overall accuracy. This study underscores the potential of using machine learning methods to enhance mammography interpretation, especially in remote areas and primary hospitals.

Pelvic CT in addition to MRI to differentiate between rectal and sigmoid cancer on imaging using the sigmoid take-off as a landmark.

The sigmoid take-off (STO) is a recently established landmark to discern rectal from sigmoid cancer on imaging. STO-assessment can be challenging on magnetic resonance imaging (MRI) due to varying axial planes. To establish the benefit of using computed tomography (CT; with consistent axial planes), in addition to MRI, to anatomically classify rectal versus sigmoid cancer using the STO. A senior and junior radiologist retrospectively classified 40 patients with rectal/rectosigmoid cancers using the STO, first on MRI-only (sagittal and oblique-axial views) and then using a combination of MRI and axial CT. Tumors were classified as rectal/rectosigmoid/sigmoid (according to published STO definitions) and then dichotomized into rectal versus sigmoid. Diagnostic confidence was documented using a 5-point scale. Adding CT resulted in a change in anatomical tumor classification in 4/40 cases (10%) for the junior reader and in 6/40 cases (15%) for the senior reader. Diagnostic confidence increased significantly after adding CT for the junior reader (mean score 3.85 vs. 4.27; P < 0.001); confidence of the senior reader was not affected (4.28 vs. 4.25; P = 0.80). Inter-observer agreement was similarly good for MRI only (κ=0.77) and MRI + CT (κ=0.76). Readers reached consensus on the classification of rectal versus sigmoid cancer in 78%-85% of cases. Availability of a consistent axial imaging plane - in the case of this study provided by CT - in addition to a standard MRI protocol with sagittal and oblique-axial imaging views can be helpful to more confidently localize tumors using the STO as a landmark, especially for more junior readers.

A deep learning algorithm to improve readers' interpretation and speed of pancreatic cystic lesions on dual-phase enhanced CT.

To develop a deep learning model (DLM) to improve readers' interpretation and speed in the differentiation of pancreatic cystic lesions (PCLs) on dual-phase enhanced CT, and a low contrast mediadose, external testing set validated the model. Dual-phase enhanced CT images of 363 patients with 368 PCLs obtained from two centers were retrospectively assessed. Based on the examination date, a training and validation set of 266 PCLs, an internal testing set of 52 PCLswere designated from center 1. Anexternal testing set included 50 PCLs from center 2. Clinical and radiological characteristics were compared. The DLM was developed using 3D specially designed densely connected convolutional networks for PCL differentiation. Radiomic features were extracted to build a traditional radiomics model (RM). Performance of the DLM, traditional RM, and three readers was compared. The accuracy for differential diagnosis was 0.904 with DLM, which was the highest in the internal testing set. Accuracy differences between the DLM and senior radiologist were not significant both in the internal and external testing set (both p > 0.05). With the help of the DLM, the accuracy and specificity of the junior radiologist were significantly improved (all p < 0.05), and all readers' diagnostic time was shortened (all p < 0.05). The DLM achieved senior radiologist-level performance in differentiating benign and malignant PCLs which could improve the junior radiologist's interpretation and speed of PCLs on CT.

Interobserver Reliability of Magnetic Resonance Imaging of Sacroiliac Joints in Axial Spondyloarthritis.

Introduction: Axial spondyloarthritis (axSpA) is characterized by damage to the axial skeleton and entheses, and is often associated with extra-articular manifestations, in the presence of the human leukocyte antigen (HLA) B27. The aim of our study is to assess the performance of rheumatologists in interpreting the inflammatory and structural damage to sacroiliac joints, in comparison to radiologists. Material and Methods: The present study included a total of 34 patients diagnosed with axSpA, according to the Assessment of SpondyloArthritis International Society (ASAS) criteria for axSpA, examined from January 2021 to November 2021 in the Departments of Rheumatology and Radiology and Medical Imaging of the University of Medicine and Pharmacy of Craiova. All patients underwent physical examination, laboratory tests, and magnetic resonance imaging (MRI) of the sacroiliac joints. The images were interpreted by a senior radiologist (SR), a junior radiologist (JR), a senior rheumatologist (SRh), and a junior rheumatologist (JRh), who were blinded to the clinical and paraclinical data. Results: The overall κ was 0.7 for the JR (substantial agreement), 0.707 for the SRh (substantial agreement), and 0.601 for the JRh (moderate agreement), in comparison with the SR. Regarding the overall inflammatory changes, the SRh and JR were proven to have substantial agreement (κ = 0.708 and 0.742, respectively) with the SR, while the JRh was proven to have moderate agreement (κ = 0.607). The structural damage observed by the JR showed substantial agreement (κ = 0.676) with the SR, while the SRh and JRh had substantial and moderate agreement (κ = 0.705 and 0.596, respectively) with the SR. Conclusions: Our study showed substantial agreement between the senior radiologist, senior rheumatologist, and junior radiologist, and moderate agreement with the junior rheumatologist.

An Artificial Intelligence Model Based on ACR TI-RADS Characteristics for US Diagnosis of Thyroid Nodules.

Background US-based diagnosis of thyroid nodules is subjective and influenced by radiologists' experience levels. Purpose To develop an artificial intelligence model based on American College of Radiology Thyroid Imaging Reporting and Data System characteristics for diagnosing thyroid nodules and identifying nodule characteristics (hereafter, MTI-RADS) and to compare the performance of MTI-RADS, radiologists, and a model trained on benign and malignant status based on surgical histopathologic analysis (hereafter, MDiag). Materials and Methods In this retrospective study, 1588 surgically proven nodules from 636 consecutive patients (mean age, 49 years ± 14 [SD]; 485 women) were included. MTI-RADS and MDiag were trained on US images of 1345 nodules (January 2018 to December 2019). The performance of MTI-RADS was compared with that of MDiag and radiologists with different experience levels on the test data set (243 nodules, January 2019 to December 2019) with the DeLong method and McNemar test. Results The area under the receiver operating characteristic curve (AUC) and sensitivity of MTI-RADS were 0.91 and 83% (55 of 66 nodules), respectively, which were not significantly different from those of experienced radiologists (0.93 [P = .45] and 92% [61 of 66 nodules; P = .07]) and exceeded those of junior radiologists (0.78 [P < .001] and 70% [46 of 66 nodules; P = .04]). The specificity of MTI-RADS (87% [154 of 177 nodules]) was higher than that of both experienced and junior radiologists (80% [141 of 177 nodules; P = .02] and 75% [133 of 177 nodules; P = .001], respectively). The AUC of MTI-RADS was higher than that of MDiag (0.91 vs 0.84, respectively; P = .001). In the test set of 243 nodules, the consistency rates between MTI-RADS and the experienced group were higher than those between MTI-RADS and the junior group for composition (79% [n = 193] vs 73% [n = 178], respectively; P = .02), echogenicity (75% [n = 183] vs 68% [n = 166]; P = .04), shape (93% [n = 227] vs 88% [n = 215]; P = .04), and smooth or ill-defined margin (72% [n = 174] vs 63% [n = 152]; P = .002). Conclusion The area under the receiver operating characteristic curve (AUC) of an artificial intelligence model based on the American College of Radiology Thyroid Imaging Reporting and Data System (TI-RADS) was higher than that of a model trained on benign and malignant status based on surgical histopathologic analysis. The AUC and sensitivity of the model based on TI-RADS exceeded those of junior radiologists; the specificity of the model was higher than that of both experienced and junior radiologists. © RSNA, 2022.

Diagnostic Accuracy of Subtraction Coronary CT Angiography in Severely Calcified Segments: Comparison Between Readers With Different Levels of Experience.

PurposeSubtraction coronary CT angiography (CCTA) may reduce blooming and beam-hardening artifacts. This study aimed to assess its value in improving the diagnostic accuracy of readers with different experience levels.MethodWe prospectively enrolled patients with target segment who underwent CCTA and invasive coronary angiography (ICA). Target segment images were independently evaluated by three groups of radiologists with different experience levels with CCTA using ICA as the standard reference. Diagnostic accuracy was measured by the area under the curve (AUC), using ≥50% stenosis as the cut-off value.ResultsIn total, 134 target segments with severe calcification from 47 patients were analyzed. The mean specificity of conventional CCTA for each group ranged from 22.4 to 42.2%, which significantly improved with subtraction CCTA, ranging from 81.3 to 85.7% (all p < 0.001). The mean sensitivity of conventional CCTA for each group ranged from 83.3 to 88.0%. Following calcification subtraction, the mean sensitivity decreased for the novice (p < 0.001) and junior (p = 0.017) radiologists but was unchanged for the senior radiologists (p = 0.690). With subtraction CCTA, the mean AUCs of CCTA significantly increased: values ranged from 0.53, 0.54, and 0.61 to 0.70, 0.74, and 0.85 for the novice, junior, and senior groups (all p < 0.001).ConclusionSubtraction CCTA could improve the diagnostic accuracy of radiologists at all experience levels of CCTA interpretation.

An artificial intelligence system using maximum intensity projection MR images facilitates classification of non-mass enhancement breast lesions.

To build an artificial intelligence (AI) system to classify benign and malignant non-mass enhancement (NME) lesions using maximum intensity projection (MIP) of early post-contrast subtracted breast MR images. This retrospective study collected 965 pure NME lesions (539 benign and 426 malignant) confirmed by histopathology or follow-up in 903 women. The 754 NME lesions acquired by one MR scanner were randomly split into the training set, validation set, and test set A (482/121/151 lesions). The 211 NME lesions acquired by another MR scanner were used as test set B. The AI system was developed using ResNet-50 with the axial and sagittal MIP images. One senior and one junior radiologist reviewed the MIP images of each case independently and rated its Breast Imaging Reporting and Data System category. The performance of the AI system and the radiologists was evaluated using the area under the receiver operating characteristic curve (AUC). The AI system yielded AUCs of 0.859 and 0.816 in the test sets A and B, respectively. The AI system achieved comparable performance as the senior radiologist (p = 0.558, p = 0.041) and outperformed the junior radiologist (p < 0.001, p = 0.009) in both test sets A and B. After AI assistance, the AUC of the junior radiologist increased from 0.740 to 0.862 in test set A (p < 0.001) and from 0.732 to 0.843 in test set B (p < 0.001). Our MIP-based AI system yielded good applicability in classifying NME lesions in breast MRI and can assist the junior radiologist achieve better performance. • Our MIP-based AI system yielded good applicability in the dataset both from the same and a different MR scanner in predicting malignant NME lesions. • The AI system achieved comparable diagnostic performance with the senior radiologist and outperformed the junior radiologist. • This AI system can assist the junior radiologist achieve better performance in the classification of NME lesions in MRI.

Multi-lesion radiomics model for discrimination of relapsing-remitting multiple sclerosis and neuropsychiatric systemic lupus erythematosus.

To develop an MRI-based multi-lesion radiomics model for discrimination of relapsing-remitting multiple sclerosis (RRMS) and its mimicker neuropsychiatric systemic lupus erythematosus (NPSLE). A total of 112 patients with RRMS (n = 63) or NPSLE (n = 49) were assigned to training and test sets with a ratio of 3:1. All lesions across the whole brain were manually segmented on T2-weighted fluid-attenuated inversion recovery images. For each single lesion, 371 radiomics features were extracted and trained using machine learning algorithms, producing Radiomics Index for Lesion (RIL) for each lesion and a single-lesion radiomics model. Then, for each subject, single lesions were assigned to one of two disease courts based on their distance to decision threshold, and a Radiomics Index for Subject (RIS) was calculated as the mean RIL value of lesions on the higher-weighted court. Accordingly, a subject-level discrimination model was constructed and compared with performances of two radiologists. The subject-based discrimination model satisfactorily differentiated RRMS and NPSLE in both training (AUC = 0.967, accuracy = 0.892, sensitivity = 0.917, and specificity = 0.872) and test sets (AUC = 0.962, accuracy = 0.931, sensitivity = 1.000, and specificity = 0.875), significantly better than the single-lesion radiomics method (training: p < 0.001; test: p = 0.001) Besides, the discrimination model significantly outperformed the senior radiologist in the training set (training: p = 0.018; test: p = 0.077) and the junior radiologist in both the training and test sets (training: p = 0.008; test: p = 0.023). The multi-lesion radiomics model could effectively discriminate between RRMS and NPSLE, providing a supplementary tool for accurate differential diagnosis of the two diseases. • Radiomic features of brain lesions in RRMS and NPSLE were different. • The multi-lesion radiomics model constructed using a merging strategy was comprehensively superior to the single-lesion-based model for discrimination of RRMS and NPSLE. • The RRMS-NPSLE discrimination model showed a significantly better performance or a trend toward significance than the radiologists.

157 Diagnostic Test Accuracy of Artificial Intelligence Models Used in Cross-Sectional Radiological Imaging of Surgical Pathology in the Abdominopelvic Cavity: A Systematic Review

Abstract Introduction Medical imaging is important for diagnostic, prognostic, and management decisions. It is reliant on an increasingly limited number of interpreters. A developing interest has explored how artificial intelligence (AI) research in medical imaging can support clinicians and provide greater efficiency in clinical care. Reviews exist for thoracic and endoscopic imaging, but one is lacking for the abdominopelvic cavity. This could benefit several specialities which use this modality of imaging to guide their clinical decision making. This systematic review examines and critically appraises the application of AI models to identify surgical pathology from cross-sectional radiological images of the abdominopelvic cavity, to identify current limitations and inform future research. Method Systematic database searches (Medline, EMBASE, Cochrane Central Register of Controlled Trials) to identify relevant studies were performed, adhering to the PRISMA-DTA guidelines. Study characteristics and outcomes assessing diagnostic performance were extracted. A narrative synthesis was performed in accordance with the Synthesis Without Meta-analysis guidelines. Results 10 retrospective studies were included, comprising 3,096 and 1,432 patients for AI training and test sets, respectively. There was diversity in the speciality, intention of the AI applications and the reporting, which was unstandardised. Diagnostic performance of models varied (range: 70–95% sensitivity, 73.7%-98% specificity). Only one study used a comparator, in which AI (AUC=0.920) outperformed both senior and junior radiologists (AUC=0.791 and 0.780, respectively). Conclusions AI application in this field is diverse and adherence to new and developing reporting guidelines is warranted. With finite healthcare resources and funding, future endeavours may benefit from prioritising clinical need, rather than scientific inquiry.

Small lesion classification on abbreviated breast MRI: training can improve diagnostic performance and inter-reader agreement.

To determine whether the diagnostic performance and inter-reader agreement for small lesion classification on abbreviated breast MRI (AB-MRI) can be improved by training, and can achieve the level of full diagnostic protocol MRI (FDP-MRI). This retrospective study enrolled 1165 breast lesions (≤ 2 cm; 409 malignant and 756 benign) from 1165 MRI examinations for reading test. Twelve radiologists were assigned into a trained group and a non-trained group. They interpreted each AB-MRI twice, which was extracted from FDP-MRI. After the first read, the trained group received a structured training for AB-MRI interpretation while the non-trained group did not. FDP-MRIs were interpreted by the trained group after the second read. BI-RADS category for each lesion was compared to the standard of reference (histopathological examination or follow-up) to calculate diagnostic accuracy. Inter-reader agreement was assessed using multirater k analysis. Diagnostic accuracy and inter-reader agreement were compared between the trained and non-trained groups, between the first and second reads, and between AB-MRI and FDP-MRI. After training, the diagnostic accuracy of AB-MRI increased from 77.6 to 84.4%, and inter-reader agreement improved from 0.410 to 0.579 (both p < 0.001), which were higher than those of the non-trained group (accuracy, 84.4% vs 78.0%; weighted k, 0.579 vs 0.461; both p < 0.001). The post-training accuracy and inter-reader agreement of AB-MRI were lower than those of FDP-MRI (accuracy, 84.4% vs 92.8%; weighted k, 0.579 vs 0.602; both p < 0.001). Training can improve the diagnostic performance and inter-reader agreement for small lesion classification on AB-MRI; however, it remains inferior to those of FDP-MRI. • Training can improve the diagnostic performance for small breast lesions on AB-MRI. • Training can reduce inter-observer variation for breast lesion classification on AB-MRI, especially among junior radiologists. • The post-training diagnostic performance and inter-reader agreement of AB-MRI remained inferior to those of FDP-MRI.

A Practical Deep Learning Model in Differentiating Pneumonia-Type Lung Carcinoma from Pneumonia on CT Images: ResNet Added with Attention Mechanism.

Objective We aim to develop a deep neural network model to differentiate pneumonia-type lung carcinoma from pneumonia based on chest CT scanning and evaluate its performance. Materials and Methods We retrospectively analyzed 131 patients diagnosed with pneumonia-type lung carcinoma and 171 patients with pneumonia treated in Beijing Hospital from October 2019 to February 2021. The average age was 68 (±15) years old, and the proportion of men (162/302) was slightly more than that of women (140/302). In this study, a deep learning based model UNet was applied to extract lesion areas from chest CT images. Lesion areas were extracted and classified by a designed spatial attention mechanism network. The model AUC and diagnostic accuracy were analyzed based on the results of the model. We analyzed the accuracy rate, sensitivity, and specificity and compared the results of the model to the junior and senior radiologists and radiologists based on the model. Results The model has a good efficiency in detecting pneumonia-like lesions (6.31 seconds/case). The model accuracy rate, sensitivity, and specificity were 74.20%, 60.37%, and 89.36%, respectively. The junior radiologist's accuracy rate, sensitivity, and specificity were 61.00%, 48.08%, and 75.00%, respectively. The senior radiologist's accuracy rate, sensitivity, and specificity were 65.00%, 51.92%, and 79.17%, respectively. The results of junior radiologists based on the model were improved (76.00% for accuracy rate, 62.75% for sensitivity, and 89.80% for specificity). The results of senior radiologists based on the model were also improved (78.00% for accuracy rate, 64.71% for sensitivity, and 91.84% for specificity) and the diagnostic accuracy of which was statistically higher than other groups (P < 0.05). Based on the lesion texture diversity and the lesion boundary ambiguity, the algorithm produced false-positive samples (13.51%). Conclusion This deep learning model could detect pneumonia-type lung carcinoma and differentiate it from pneumonia accurately and efficiently.