Real-World Experience with Artificial Intelligence-Based Triage in Transferred Large Vessel Occlusion Stroke Patients.

BACKGROUND AND PURPOSE: Randomized controlled trials have demonstrated the importance of time to endovascular therapy (EVT) in clinical outcomes in large vessel occlusion (LVO) acute ischemic stroke. Delays to treatment are particularly prevalent when patients require a transfer from hospitals without EVT capability onsite. A computer-aided triage system, Viz LVO, has the potential to streamline workflows. This platform includes an image viewer, a communication system, and an artificial intelligence (AI) algorithm that automatically identifies suspected LVO strokes on CTA imaging and rapidly triggers alerts. We hypothesize that the Viz application will decrease time-to-treatment, leading to improved clinical outcomes. METHODS: A retrospective analysis of a prospectively maintained database was assessed for patients who presented to a stroke center currently utilizing Viz LVO and underwent EVT following transfer for LVO stroke between July 2018 and March 2020. Time intervals and clinical outcomes were compared for 55 patients divided into pre- and post-Viz cohorts. RESULTS: The median initial door-to-neuroendovascular team (NT) notification time interval was significantly faster (25.0 min [IQR = 12.0] vs. 40.0 min [IQR = 61.0]; p = 0.01) with less variation (p < 0.05) following Viz LVO implementation. The median initial door-to-skin puncture time interval was 25 min shorter in the post-Viz cohort, although this was not statistically significant (p = 0.15). CONCLUSIONS: Preliminary results have shown that Viz LVO implementation is associated with earlier, more consistent NT notification times. This application can serve as an early warning system and a failsafe to ensure that no LVO is left behind.

Association of Coronavirus Disease (COVID-19) With Large Vessel Occlusion Strokes: A Case-Control Study.

BACKGROUND. An increase in frequency of acute ischemic strokes has been observed among patients presenting with acute neurologic symptoms during the coronavirus disease (COVID-19) pandemic. OBJECTIVE. The purpose of this study was to investigate the association between COVID-19 and stroke subtypes in patients presenting with acute neurologic symptoms. METHODS. This retrospective case-control study included patients for whom a code for stroke was activated from March 16 to April 30, 2020, at any of six New York City hospitals that are part of a single health system. Demographic data (age, sex, and race or ethnicity), COVID-19 status, stroke-related risk factors, and clinical and imaging findings pertaining to stroke were collected. Univariate and multivariate analyses were conducted to evaluate the association between COVID-19 and stroke subtypes. RESULTS. The study sample consisted of 329 patients for whom a code for stroke was activated (175 [53.2%] men, 154 [46.8%] women; mean age, 66.9 ± 14.9 [SD] years). Among the 329 patients, 35.3% (116) had acute ischemic stroke confirmed with imaging; 21.6% (71) had large vessel occlusion (LVO) stroke; and 14.6% (48) had small vessel occlusion (SVO) stroke. Among LVO strokes, the most common location was middle cerebral artery segments M1 and M2 (62.0% [44/71]). Multifocal LVOs were present in 9.9% (7/71) of LVO strokes. COVID-19 was present in 38.3% (126/329) of the patients. The 61.7% (203/329) of patients without COVID-19 formed the negative control group. Among individual stroke-related risk factors, only Hispanic ethnicity was significantly associated with COVID-19 (38.1% of patients with COVID-19 vs 20.7% of patients without COVID-19; p = 0.001). LVO was present in 31.7% of patients with COVID-19 compared with 15.3% of patients without COVID-19 (p = 0.001). SVO was present in 15.9% of patients with COVID-19 and 13.8% of patients without COVID-19 (p = 0.632). In multivariate analysis controlled for race and ethnicity, presence of COVID-19 had a significant independent association with LVO stroke (odds ratio, 2.4) compared with absence of COVID-19 (p = 0.011). CONCLUSION. COVID-19 is associated with LVO strokes but not with SVO strokes. CLINICAL IMPACT. Patients with COVID-19 presenting with acute neurologic symptoms warrant a lower threshold for suspicion of large vessel stroke, and prompt workup for large vessel stroke is recommended.

Estimation of Ischemic Core Volume Using Computed Tomographic Perfusion.

Background and Purpose- Estimation of infarction based on computed tomographic perfusion (CTP) has been challenging, mainly because of noise associated with CTP data. The Bayesian method is a robust probabilistic method that minimizes effects of oscillation, tracer delay, and noise during residue function estimation compared with other deconvolution methods. This study compares CTP-estimated ischemic core volume calculated by the Bayesian method and by the commonly used block-circulant singular value deconvolution technique. Methods- Patients were included if they had (1) anterior circulation ischemic stroke, (2) baseline CTP, (3) successful recanalization defined by thrombolysis in cerebral infarction ≥IIb, and (4) minimum infarction volume of >5 mL on follow-up magnetic resonance imaging (MRI). CTP data were processed with circulant singular value deconvolution and Bayesian methods. Two established CTP methods for estimation of ischemic core volume were applied: cerebral blood flow (CBF) method (relative CBF, <30% within the region of delay >2 seconds) and cerebral blood volume method (<2 mL per 100 g within the region of relative mean transit time >145%). Final infarct volume was determined on MRI (fluid-attenuated inversion recovery images). CTP and MRI-derived ischemic core volumes were compared by univariate and Bland-Altman analysis. Results- Among 35 patients included, the mean/median (mL) difference for CTP-estimated ischemic core volume against MRI was -4/-7 for Bayesian CBF ( P=0.770), 20/12 for Bayesian cerebral blood volume ( P=0.041), 21/10 for circulant singular value deconvolution CBF ( P=0.006), and 35/18 for circulant singular value deconvolution cerebral blood volume ( P<0.001). Among all methods, Bayesian CBF provided the narrowest limits of agreement (-28 to 19 mL) in comparison with MRI. Conclusions- Despite existing variabilities between CTP postprocessing methods, Bayesian postprocessing increases accuracy and limits variability in CTP estimation of ischemic core.

Interval Change in Diffusion and Perfusion MRI Parameters for the Assessment of Pseudoprogression in Cerebral Metastases Treated With Stereotactic Radiation.

OBJECTIVE: Apparent increases in the size of cerebral metastases after stereotactic radiosurgery (SRS) can be caused by pseudoprogression or true disease progression, which poses a diagnostic challenge at conventional MRI. The purpose of this study was to assess whether interval change in DWI and perfusion MRI parameters can differentiate pseudoprogression from progressive disease after treatment with SRS. MATERIALS AND METHODS: Patients with apparent growth of cerebral metastases after SRS treatment who underwent pre- and post-SRS DWI, dynamic susceptibility contrast (DSC)-MRI, and perfusion dynamic contrast-enhanced (DCE)-MRI were retrospectively evaluated. Final assignment of pseudoprogression or progressive disease was determined at 6-month follow-up imaging using the Response Assessment in Neuro-Oncology Brain Metastases criteria. Mean values of apparent diffusion coefficient (ADC), DCE-MRI-derived volume transfer constant (Ktrans), and DSC-MRI-derived relative cerebral blood volume (CBV) from pre- and post-SRS MRI scans were compared between groups using univariate and regression analysis. Fisher exact test was used to compare interval change of imaging biomarkers. RESULTS: Of 102 cerebral metastases evaluated, 32 lesions in 29 patients met our inclusion criteria. The mean duration of follow-up was 7.2 months (range, 6-14 months). Twenty-two lesions were determined as pseudoprogression, and 10 lesions were determined as progressive disease using the Response Assessment in Neuro-Oncology Brain Metastases criteria at 6-month follow-up MRI. The interval change pattern of our imaging parameters matched the expected patterns of treatment response for ADC (23/32 lesions; 72%; p = 0.055; odds ratio, 5.1), Ktrans (24/32 lesions; 75%; p = 0.006; odds ratio, 19.2), and relative CBV (27/32 lesions; 84%; p = 0.001; odds ratio, 25.3). CONCLUSION: Pseudoprogression can be distinguished from disease progression in cerebral metastases treated with SRS via an interval decrease in relative CBV and Ktrans values.

Neuroimaging in Central Nervous System Infections.

Radiologic imaging is often employed to supplement clinical evaluation in cases of suspected central nervous system (CNS) infection. While computed tomography (CT) is superior for evaluating osseous integrity, demineralization, and erosive changes and may be more readily available at many institutions, magnetic resonance imaging (MRI) has significantly greater sensitivity for evaluating the cerebral parenchyma, cord, and marrow for early changes that have not yet reached the threshold for CT detection. For these reasons, MRI is generally superior to CT for characterizing bacterial, viral, fungal, and parasitic infections of the CNS. The typical imaging features of common and uncommon CNS infectious processes are reviewed.

Tissue-Negative Transient Ischemic Attack: Is There a Role for Perfusion MRI?

OBJECTIVE: Approximately 60% of patients with a clinical transient ischemic attack (TIA) do not have DWI evidence of cerebral ischemia. The purpose of this study was to assess the added diagnostic value of perfusion MRI in the evaluation of patients with TIA who have normal DWI findings. MATERIALS AND METHODS: The inclusion criteria for this retrospective study were clinical presentation of TIA at admission with a discharge diagnosis of TIA confirmed by a stroke neurologist, MRI including both DWI and perfusion-weighted imaging within 48 hours of symptom onset, and no DWI lesion. Cerebral blood flow (CBF) and time to maximum of the residue function (Tmax) maps were evaluated independently by two observers. Multivariate analysis was used to assess perfusion findings; clinical variables; age, blood pressure, clinical symptoms, diabetes (ABCD2) score; duration of TIA; and time between MRI and onset and resolution of symptoms. RESULTS: Fifty-two patients (33 women, 19 men; age range, 20-95 years) met the inclusion criteria. A regional perfusion abnormality was identified on either Tmax or CBF maps of 12 of 52 (23%) patients. Seven (58%) of the patients with perfusion abnormalities had hypoperfused lesions best detected on Tmax maps; the other five had hyperperfusion best detected on CBF maps. In 11 of 12 (92%) patients with abnormal perfusion MRI findings, the regional perfusion deficit correlated with the initial neurologic deficits. Multivariable analysis revealed no significant difference in demographics, ABCD2 scores, or presentation characteristics between patients with and those without perfusion abnormalities. CONCLUSION: Perfusion MRI that includes Tmax and CBF parametric maps adds diagnostic value by depicting regions with delayed perfusion or postischemic hyperperfusion in approximately one-fourth of TIA patients who have normal DWI findings.

Effects of Liposomal Bupivacaine With Adductor Canal Block on Pain and Functional Outcomes in Patients Undergoing Total Knee Arthroplasty.

BACKGROUND: Liposomal bupivacaine (LB) has an extended duration of action compared to bupivacaine and may allow patients to reach physical therapy (PT) goals faster than traditional methods. OBJECTIVE: To determine the effect of a periarticular LB mixture with adductor canal blockade (ACB) on postoperative opioid requirements, pain scores, and functional outcomes in patients undergoing primary unilateral total knee arthroplasty. METHODS: A retrospective chart review was performed; 86 patients received the LB mixture + ACB, and 86 historical controls received nonliposomal bupivacaine and femoral nerve block (FNB). RESULTS: There was no effect of group on mean postoperative pain scores (P = 0.144). There was an effect of group on equivalent morphine dose (P = 0.008). Pain scores and morphine doses changed over time in both groups, but there was no time-group interaction. Compared with controls, patients in the LB mixture group were more likely to require minimal assistance or better when going from a sitting to a standing position by postoperative day 1 (POD; 99% vs 81%, P = 0.0001) and POD 2 (90% vs 77%, P = 0.0212). There were no differences between groups with regard to discharge disposition or safety outcomes. Use did not affect LOS for patients discharged to an extended care facility but did reduce LOS for those discharged home. CONCLUSIONS: The LB mixture was effective in reducing opioid use and improving functional outcomes compared with historical controls.

Contrast-Enhanced Dixon Fat-Water Separation Imaging of the Spine: Added Value of Fat, In-Phase and Opposed-Phase Imaging in Marrow Lesion Detection.

OBJECTIVE: Beyond fat suppression (FS), the efficacy of (fat-water separation or Dixon [FWD]) Dixon imaging in gadolinium-enhanced spine imaging has yet to be validated. This study evaluated enhanced opposed-phase (OP) and fat-only (FO) images along with water-only (WO; FS) images against traditional unenhanced techniques and rated the incremental value of in-phase imaging in patients with presumed neoplastic focal spine lesions. METHODS: A retrospective cohort study of 36 subjects with focal spine lesions imaged with FWD was evaluated qualitatively and quantitatively. RESULTS: Enhanced OP, WO, and FO images were of significant value in detection of osseous lesions, surpassing the lesion conspicuity with conventional techniques both qualitatively and quantitatively, although the impact of in-phase imaging was limited. Water-only imaging performed well for FS. CONCLUSIONS: Contrast-enhanced FO, WO, and OP outperform traditional techniques, providing reliable lesion characterization and highest conspicuity. In-phase imaging offered limited impact on the subjective assessment of enhancement. The added value and robustness of FWD, particularly the unique contrast provided by FO imaging, suggests consideration for routine use for postgadolinium spine imaging.

CT of Anatomic Variants of the Paranasal Sinuses and Nasal Cavity: Poor Correlation With Radiologically Significant Rhinosinusitis but Importance in Surgical Planning.

OBJECTIVE: The purpose of this study was to determine the incidence of sinonasal anatomic variants and to assess their relation to sinonasal mucosal disease. MATERIALS AND METHODS: A retrospective evaluation of 192 sinus CT examinations of patients with a clinical history of rhinosinusitis was conducted. The CT scans were evaluated for the presence of several anatomic variants of the sinonasal cavities, and the prevalence of each variant was calculated. Prevalences of all sinonasal anatomic variants were compared between patients who had minimal to no apparent imaging evidence of rhinosinusitis and those who had radiologic evidence of clinically significant rhinosinusitis. RESULTS: The most common normal variants were nasal septal deviation, Agger nasi cells, and extension of the sphenoid sinuses into the posterior nasal septum. We found no statistically significant difference in the prevalence of any of the studied anatomic variants between patients with minimal and those with clinically significant paranasal sinus or nasal cavity disease. CONCLUSION: Analysis of every routine CT scan of the paranasal sinuses obtained for sinusitis or rhinitis for the presence of different anatomic variants is of questionable value unless surgery is planned.

Impact of Deep Learning Image Reconstruction Methods on MRI Throughput.

Purpose To evaluate the effect of implementing two distinct commercially available deep learning reconstruction (DLR) algorithms on the efficiency of MRI examinations conducted in real clinical practice within an outpatient setting at a large, multicenter institution. Materials and Methods This retrospective study included 7346 examinations from 10 clinical MRI scanners analyzed during the pre- and postimplementation periods of DLR methods. Two different types of DLR methods, namely Digital Imaging and Communications in Medicine (DICOM)-based and k-space-based methods, were implemented in half of the scanners (three DICOM-based and two k-space-based), while the remaining five scanners had no DLR method implemented. Scan and room times of each examination type during the pre- and postimplementation periods were compared among the different DLR methods using the Wilcoxon test. Results The application of deep learning methods resulted in significant reductions in scan and room times for certain examination types. The DICOM-based method demonstrated up to a 53% reduction in scan times and a 41% reduction in room times for various study types. The k-space-based method demonstrated up to a 27% reduction in scan times but did not significantly reduce room times. Conclusion DLR methods were associated with reductions in scan and room times in a clinical setting, though the effects were heterogeneous depending on examination type. Thus, potential adopters should carefully evaluate their case mix to determine the impact of integrating these tools. Keywords: Deep Learning MRI Reconstruction, Reconstruction Algorithms, DICOM-based Reconstruction, k-Space-based Reconstruction © RSNA, 2024 See also the commentary by GharehMohammadi in this issue.

Academic Neuroradiology: 2023 Update on Turn-Around Time, Financial Recruitment and Retention Strategies.

The ASNR Neuroradiology Division Chief Working Group's 2023 survey, with responses from 62 division chiefs, provides insights into turn-around times, faculty recruitment, moonlighting opportunities, and academic funds.In emergency cases, 61% aim for a turn-around time of less than 45-60 minutes, with two-thirds meeting this expectation more than 75% of the time. For inpatient CT and MRI scans, 54% achieve a turn-around time of 4-8 hours, with three quarters meeting this expectation at least 50% of the time. Outpatient scans have an expected turn-around time of 24-48 hours, which is met in 50% of cases.Faculty recruitment strategies included 35% offering sign-on bonuses, with a median of $30,000. Additionally, 23% provided bonuses to fellows during fellowship to retain them in the practice upon completion of their fellowship. Internal moonlighting opportunities for faculty were offered by 70% of divisions, with a median pay of $250 per hour.The median annual academic fund for a full-time neuroradiology faculty member was $6,000, typically excluding license fees but including ACR and ABR membership, leaving $4,000 for professional expenses.This survey calls for further dialogue on adapting and innovating academic institutions to meet evolving needs in neuroradiology.

Diffusion Weighted Imaging in Spine Tumors.

Diffusion weighted imaging (DWI) has developed into a powerful tool for the evaluation of spine tumors, particularly for the assessment of vertebral marrow lesions and intramedullary tumors. Advances in magnetic resonance techniques have improved the quality of spine DWI and diffusion tensor imaging (DTI) in recent years, with increased reproducibility and utilization. DTI, with quantitative parameters such as fractional anisotropy and qualitative visual assessment of nerve fiber tracts, can play a valuable role in the evaluation and surgical planning of spinal cord tumors. These widely available techniques can be used to enhance the diagnostic evaluation of spinal tumors.

AI software detection of large vessel occlusion stroke on CT angiography: a real-world prospective diagnostic test accuracy study.

BACKGROUND: Artificial intelligence (AI) software is increasingly applied in stroke diagnostics. However, the actual performance of AI tools for identifying large vessel occlusion (LVO) stroke in real time in a real-world setting has not been fully studied. OBJECTIVE: To determine the accuracy of AI software in a real-world, three-tiered multihospital stroke network. METHODS: All consecutive head and neck CT angiography (CTA) scans performed during stroke codes and run through an AI software engine (Viz LVO) between May 2019 and October 2020 were prospectively collected. CTA readings by radiologists served as the clinical reference standard test and Viz LVO output served as the index test. Accuracy metrics were calculated. RESULTS: Of a total of 1822 CTAs performed, 190 occlusions were identified; 142 of which were internal carotid artery terminus (ICA-T), middle cerebral artery M1, or M2 locations. Accuracy metrics were analyzed for two different groups: ICA-T and M1 ±M2. For the ICA-T/M1 versus the ICA-T/M1/M2 group, sensitivity was 93.8% vs 74.6%, specificity was 91.1% vs 91.1%, negative predictive value was 99.7% vs 97.6%, accuracy was 91.2% vs 89.8%, and area under the curve was 0.95 vs 0.86, respectively. Detection rates for ICA-T, M1, and M2 occlusions were 100%, 93%, and 49%, respectively. As expected, the algorithm offered better detection rates for proximal occlusions than for mid/distal M2 occlusions (58% vs 28%, p=0.03). CONCLUSIONS: These accuracy metrics support Viz LVO as a useful adjunct tool in stroke diagnostics. Fast and accurate diagnosis with high negative predictive value mitigates missing potentially salvageable patients.

Neuroimaging findings in 4342 hospitalized COVID-19 subjects: A multicenter report from the United States and Europe.

BACKGROUND AND PURPOSE: To determine the incidence of acute neuroimaging (NI) findings and comorbidities in the coronavirus disease of 2019 (COVID-19)-infected subjects in seven U.S. and four European hospitals. METHODS: This is a retrospective study of COVID-19-positive subjects with the following inclusion criteria: age >18, lab-confirmed COVID-19 infection, and acute NI findings (NI+) attributed to COVID-19 on CT or MRI brain. NI+ and comorbidities in total hospitalized COVID-19-positive (TN) subjects were assessed. RESULTS: A total of 37,950 COVID-19-positive subjects were reviewed and 4342 underwent NI. NI+ incidence in subjects with NI was 10.1% (442/4342) including 7.9% (294/3701) in the United States and 22.8% (148/647) in Europe. NI+ incidence in TN was 1.16% (442/37,950). In NI (4342), incidence of ischemic stroke was 6.4% followed by intracranial hemorrhage (ICH) (3.8%), encephalitis (0.5%), sinus venous thrombosis (0.2%), and acute disseminated encephalomyelitis (ADEM) (0.2%). White matter involvement was seen in 57% of NI+. Hypertension was the most common comorbidity (54%) before cardiac disease (28.8%) and diabetes mellitus (27.7%). Cardiac disease (p < .025), diabetes (p < .014), and chronic kidney disease (p < .012) were more common in the United States. CONCLUSION: This multicenter, multinational study investigated the incidence and spectrum of NI+ in 37,950 hospitalized adult COVID-19 subjects including regional differences in incidences of NI+, associated comorbidities, and other demographics. NI+ incidence in TN was 1.16% including 0.95% in the United States and 2.09% in Europe. ICH, encephalitis, and ADEM were common in Europe, while ischemic strokes were more common in the United States. In this cohort, incidence and distribution of NI+ helped characterize the neurological complications of COVID-19.

Automated assessment of ischemic core on non-contrast computed tomography: a multicenter comparative analysis with CT perfusion.

BACKGROUND: Application of machine learning (ML) algorithms has shown promising results in estimating ischemic core volumes using non-contrast CT (NCCT). OBJECTIVE: To assess the performance of the e-Stroke Suite software (Brainomix) in assessing ischemic core volumes on NCCT compared with CT perfusion (CTP) in patients with acute ischemic stroke. METHODS: In this retrospective multicenter study, patients with anterior circulation large vessel occlusions who underwent pretreatment NCCT and CTP, successful reperfusion (modified Thrombolysis in Cerbral Infarction ≥2b), and post-treatment MRI, were included from three stroke centers. Automated calculation of ischemic core volumes was obtained on NCCT scans using ML algorithm deployed by e-Stroke Suite and from CTP using Olea software (Olea Medical). Comparative analysis was performed between estimated core volumes on NCCT and CTP and against MRI calculated final infarct volume (FIV). RESULTS: A total of 111 patients were included. Estimated ischemic core volumes (mean±SD, mL) were 20.4±19.0 on NCCT and 19.9±18.6 on CTP, not significantly different (P=0.82). There was moderate (r=0.40) and significant (P<0.001) correlation between estimated core on NCCT and CTP. The mean difference between FIV and estimated core volume on NCCT and CTP was 29.9±34.6 mL and 29.6±35.0 mL, respectively (P=0.94). Correlations between FIV and estimated core volume were similar for NCCT (r=0.30, P=0.001) and CTP (r=0.36, P<0.001). CONCLUSIONS: Results show that ML-based estimated ischemic core volumes on NCCT are comparable to those obtained from concurrent CTP in magnitude and in degree of correlation with MR-assessed FIV.

U-Net Based Segmentation and Characterization of Gliomas.

(1) Background: Gliomas are the most common primary brain neoplasms accounting for roughly 40−50% of all malignant primary central nervous system tumors. We aim to develop a deep learning-based framework for automated segmentation and prediction of biomarkers and prognosis in patients with gliomas. (2) Methods: In this retrospective two center study, patients were included if they (1) had a diagnosis of glioma with known surgical histopathology and (2) had preoperative MRI with FLAIR sequence. The entire tumor volume including FLAIR hyperintense infiltrative component and necrotic and cystic components was segmented. Deep learning-based U-Net framework was developed based on symmetric architecture from the 512 × 512 segmented maps from FLAIR as the ground truth mask. (3) Results: The final cohort consisted of 208 patients with mean ± standard deviation of age (years) of 56 ± 15 with M/F of 130/78. DSC of the generated mask was 0.93. Prediction for IDH-1 and MGMT status had a performance of AUC 0.88 and 0.62, respectively. Survival prediction of <18 months demonstrated AUC of 0.75. (4) Conclusions: Our deep learning-based framework can detect and segment gliomas with excellent performance for the prediction of IDH-1 biomarker status and survival.

RadImageNet: An Open Radiologic Deep Learning Research Dataset for Effective Transfer Learning.

Purpose: To demonstrate the value of pretraining with millions of radiologic images compared with ImageNet photographic images on downstream medical applications when using transfer learning. Materials and Methods: This retrospective study included patients who underwent a radiologic study between 2005 and 2020 at an outpatient imaging facility. Key images and associated labels from the studies were retrospectively extracted from the original study interpretation. These images were used for RadImageNet model training with random weight initiation. The RadImageNet models were compared with ImageNet models using the area under the receiver operating characteristic curve (AUC) for eight classification tasks and using Dice scores for two segmentation problems. Results: The RadImageNet database consists of 1.35 million annotated medical images in 131 872 patients who underwent CT, MRI, and US for musculoskeletal, neurologic, oncologic, gastrointestinal, endocrine, abdominal, and pulmonary pathologic conditions. For transfer learning tasks on small datasets-thyroid nodules (US), breast masses (US), anterior cruciate ligament injuries (MRI), and meniscal tears (MRI)-the RadImageNet models demonstrated a significant advantage (P < .001) to ImageNet models (9.4%, 4.0%, 4.8%, and 4.5% AUC improvements, respectively). For larger datasets-pneumonia (chest radiography), COVID-19 (CT), SARS-CoV-2 (CT), and intracranial hemorrhage (CT)-the RadImageNet models also illustrated improved AUC (P < .001) by 1.9%, 6.1%, 1.7%, and 0.9%, respectively. Additionally, lesion localizations of the RadImageNet models were improved by 64.6% and 16.4% on thyroid and breast US datasets, respectively. Conclusion: RadImageNet pretrained models demonstrated better interpretability compared with ImageNet models, especially for smaller radiologic datasets.Keywords: CT, MR Imaging, US, Head/Neck, Thorax, Brain/Brain Stem, Evidence-based Medicine, Computer Applications-General (Informatics) Supplemental material is available for this article. Published under a CC BY 4.0 license.See also the commentary by Cadrin-Chênevert in this issue.