Articles on the development of medical image artificial intelligence (AI) algorithms are numerous in the literature, but deployment to clinical practice is infrequently discussed. The Enterprise Radiology Framework for AI Software Technology Team at Mayo Clinic has been focused on bridging the gap in clinical translation of medical image AI algorithms since its inception in 2019. During this time, we have released 17 algorithms into our radiology clinical practice. Recently, we have placed an increased focus on monitoring these algorithms, as there are few reports with practical experience documented in the literature. Our increased monitoring efforts include daily, weekly, and yearly monitoring of utilization, failure modes, data drift, and end-user feedback through automated alerts, dedicated dashboards, and pointed investigations to enable optimal algorithmic processing. End-user feedback is elicited yearly during annual reviews to ensure clinical needs are still being met. Automated monitoring has enabled earlier identification of problems, such as images no longer routing through the orchestration engine to the appropriate algorithm, minimizing potential disruption to the clinical practice and ensuring continued algorithmic utilization. Monitoring has also reinforced the importance of key aspects of interdisciplinary research and translation, such as early discussions on clinical needs coupled with technological ability and proper training. By providing our experience in and continuing to improve monitoring methods as a community, we can all minimize risk and maximize the benefits of medical pixel-based AI.

BACKGROUND AND PURPOSE:3D segmentation and volumetry of vestibular schwannomas (VSs) is a more accurate method to determine tumor growth on serial imaging, but manual annotation is time-consuming to implement in routine clinical practice. We evaluated and compared 5 deep learning-based segmentation models (nnUNet [Base, ResEncL], U-Mamba, UNETR, and MedSAM) for 3D VS segmentation and volumetry, and we examined the robustness to acquisition heterogeneity and generalization on an external cohort. MATERIALS AND METHODS:Our refined internal data set consisted of T1-contrast-enhanced images, including 2692 scans (n =383 patients) for training and 277 scans (n = 97 patients) for testing. Post-model training and validation, performance was evaluated on both internal and a publicly available external test set (n = 241) using the Dice similarity coefficient, maximum distance between the predicted and ground truth boundaries (Hausdorff distance), surface-to-surface (S2S) distance, and relative volume error (RVE). A subanalysis of the model performance was also performed to evaluate the impact of tumor volumes and data set heterogeneity. RESULTS:The median Dice score on the external test set varied between 0.899 and 0.927 with U-Mamba achieving the highest performance, followed by nnUNet (Base and ResEncL). For these top 3 models, the median Hausdorff distance was 3.59 mm, while the 95th percentile Hausdorff distance was 1.6 mm. The S2S distance was <1 mm, and the median RVE (%) varied between 0.07 and 0.08. The median Dice scores were lower, 0.848-0.85, for smaller tumors (<200 mm3) and higher for tumors of >400 mm3 (median Dice score, 0.925-0.932). CONCLUSIONS:Models based on convolutional neural networks, transformer networks, and foundational models show robust performance for VS segmentation. Given the consistently high performance and self-optimizing frameworks of convolutional neural network-based models (U-Mamba, nnUNet), these may be more suitable for clinical applications.

BACKGROUND:The optic pathway is a complex neural structure responsible for transmitting visual information from the retina to the brain. Traditionally, the optic pathway has been depicted using two-dimensional (2D) illustrations, which, while useful for simplification, can obscure depth, orientation, and connectivity, limiting a full understanding of its three-dimensional (3D) nature which is important for surgical planning and neuroanatomy education. Due to a convergence of advancing technologies in MRI image acquisition, medical CAD and 3D illustration software, as well as 3D printing technologies, these 3D visualizations can now be physically manufactured to provide life size, patient specific, physical, color-coded 3D models. 3D models manufactured from advanced imaging can provide a more accurate, interactive, non-invasive, cost-effective alternative to medical illustration and animation than traditional dissected cadaveric anatomical specimens for both clinical and educational purposes. METHODS:The source data for this project came from both a 42 year old male patient and a 21 year old male volunteer after both had been scanned on the same seven tesla MRI including DTI for the patient and volumetric sequences for the volunteer. The model was created by segmenting the optic pathway using medical CAD software and 3D illustration software. The DTI tracts were coregistered to the anatomic brain. The model was optimized for printing and hypothetical "lesions" were added along the pathway with their corresponding visual deficits. The model was printed on an HP580 multijet fusion color printer and photorealistic eyes were printed using material jetting of photopolymer via a Stratasys J750 printer. RESULTS:Multiple challenges were overcome to successfully create a life size, physical, multicolor 3D printed representation of the optic pathway created from 7T MRI data. CONCLUSION:This workflow resulted in a unique educational 3D representation of the human optic pathway that allows for direct manipulation, haptic feedback, and clear understanding of the anatomic relations both of this system normally and the correlations between lesion location and resultant expected visual field impairment. As opposed to the inconvenience, costs, and limited access accompanying the classical standard of advanced dissections of human specimens, this model is available to all learners in all environments.

Integration of AI-enabled algorithms into the radiology workflow presents a complex array of challenges that span operational, technical, clinical, and regulatory domains. Successfully overcoming these hurdles requires a multifaceted approach, including strategic planning, educational initiatives, and careful consideration of the practical implications for radiologists' workloads. Institutions must navigate these challenges with a clear understanding of the potential benefits and limitations of both vended and in-house developed AI tools.

Exam protocoling is a significant non-interpretive task burden for radiologists. The purpose of this work was to develop a natural language processing (NLP) artificial intelligence (AI) solution for automated protocoling of standard abdomen and pelvic magnetic resonance imaging (MRI) exams from basic associated order information and patient metadata. This Institutional Review Board exempt retrospective study used de-identified metadata from consecutive adult abdominal and pelvic MRI scans performed at our institution spanning 2.5 years from 2019 to 2021 to fine-tune an AI model to predict the exam protocol. The NLP algorithm Bidirectional Encoder Representations from Transformers (BERT) was employed in sequence classification mode. Twelve months of data from the COVID pandemic were excluded to avoid bias from known practice and referral pattern disruptions, with approximately 46,000 MRI exams in the resulting cohort. The final trained model had an accuracy of 88.5% with a Matthews correlation coefficient of 0.874, a true positive rate of 0.872, and a true negative rate of 0.995. Subsequent expert review of the errors performed to satisfy departmental leadership showed 81.9% were in fact correct or reasonable alternative protocols, yielding real-world performance accuracy of 97.9%. We conclude that NLP algorithms, including "smaller" large language models like the BERT family often overlooked today, can predict MRI imaging protocols for the abdomen and pelvis with high real-world performance, offering to decrease radiologists' non-interpretive task load and increasing departmental efficiency.

BACKGROUND AND PURPOSE:7T MRI is a promising clinical technology for epilepsy imaging. Quantification of hippocampus volume on MRI is a clinically useful biomarker in epilepsy. Applying automated hippocampus volume measurement tools to 7T MRI is needed to optimize the use of clinical ultra-high-field strength epilepsy imaging. The objective of this study is a performance evaluation of automated hippocampal volume measurement software at 7T MRI in both normal participants and those with seizure disorders. MATERIALS AND METHODS:7T MRI examinations were prospectively acquired in 50 participants. A subset of participants also underwent 3T MRI examinations, and a subset underwent 2 separate 7T acquisitions. Automated segmentation of the hippocampus was performed with 2 commonly used software packages (FreeSurfer and NeuroQuant) at 3T and 7T, with hippocampal volumes calculated for segmentations without any visually unacceptable errors as determined by radiologist review. Hippocampal volumes were also measured from manual segmentations, and the intraclass correlation coefficient (ICC) was used to compare data with automated segmentation volumes. RESULTS:Visually unacceptable automated hippocampus segmentation errors occurred more frequently at 7T than at 3T with NeuroQuant (11.0% versus 7.14%) and FreeSurfer (12.5% versus 0%). Computerized volume measurements at 7T correlated poorly with manual segmentation for both software programs (ICC <0.4). Hippocampal volume estimate correlation between matched 7T and 3T MRI in the same participant was fair (ICC = 0.4-0.59) to good (0.6-0.75) for software and manual segmentation. For repeated 7T MRI examinations in the same participant, hippocampus segmentation reproducibility was excellent (0.75) for automated software but poor (< 0.4) for manual segmentation. CONCLUSIONS:Computerized volume measurement of the hippocampus at 7T correlates poorly with volumes obtained through manual segmentation and suboptimally with matched 3T examination measurements, but is highly reproducible at 7T within the same participant. Segmentation errors are more common with 7T examinations, and further development of a hippocampal segmentation method specific to 7T MRI is needed to fully realize the benefits of 7T MRI for imaging patients with epilepsy.

Purpose:We aimed to use a validated artificial intelligence (AI) algorithm to extract muscle and adipose areas from CT images before radical cystectomy (RCx) and then correlate these measures with 90-day post-RCx complications. Materials and Methods:A tertiary referral center's cystectomy registry was queried for patients who underwent RCx between 2009 and 2017 for bladder cancer. Eight hundred forty-three RCx patients with CT imaging within 90 days of preceding surgery were included, to allow for extraction of body composition parameters by AI. We assessed complications within 90 days of surgery including wound, infectious, and major complications; readmission; and death. Multivariable logistic regressions associated pre-RCx measures with post-RCx complications. Results:Increasing subcutaneous adipose tissue was associated with more wound complications, while patients with increasing visceral adipose tissue had greater odds of infectious-related complications. After adjusting for patient characteristics, every 10 cm2 increases in fat mass index were associated with more infectious (odds ratio [OR], 1.04; P = .002) and wound (OR, 1.06; P < .001) complications. On multivariable analysis, a higher preoperative skeletal muscle index was associated with lower odds of major complications (OR, 0.75 for every 10 cm2; P = .008), while higher intramuscular adipose was associated with higher odds of major complications (OR, 1.93; P = .008). Conclusions:Automated AI body composition measurements preoperatively are associated with post-RCx complications. These measurements, in addition to patient (Eastern Cooperative Oncology Group performance status and smoking status) and surgical (robotic approach and continent diversion) characteristics, can then be used to individualize patient counseling and facilitate triage of nutritional and rehabilitation efforts.

Background: Sarcopenia, characterized by low muscle mass, and aberrant adiposity changes, including visceral fat accumulation, has been associated with impaired physiologic stress response and wound healing. Artificial urinary sphincter (AUS) placement is the preferred surgical treatment for men with severe post-prostatectomy incontinence. Given the higher rates of maladaptive body composition changes in this older, high comorbidity population, this study explores their impact on AUS outcomes. Methods: A retrospective analysis was performed including men who underwent primary AUS placement at the Mayo Clinic from 1999 to 2023 for post-prostatectomy incontinence and had cross sectional imaging available within 12 months prior to AUS implant. Sarcopenia and body composition were assessed from the available computed tomography (CT) scan using an algorithm that measures the area of different tissues at the L3 abdominal cross-section. The study investigated the association between sarcopenia [defined as skeletal muscle index (SMI) <52.4 cm(2)/m(2)] and adiposity (defined by total visceral and subcutaneous fat area) with all-cause reoperation, including specific etiologies of device infection/erosion, urethral atrophy, and device malfunction, using Cox proportional hazards models. Results: There were 111 patients who had available imaging within the study timeframe, 61 (55%) of whom were classified as sarcopenic. Follow-up did not differ significantly between the two groups [2.11 (0.53-4.78) vs. 2.52 (0.36-5.80) years, P=0.52]. Sarcopenic patients had a lower body mass index (BMI) (29.1 vs. 32.7 kg/m(2); P<0.001). No significant difference in overall device survival was observed between sarcopenic and non-sarcopenic patients (P=0.94) on Cox survival analysis. Sarcopenic patients had higher device infection rates, accounting for 16.7% (3/18) of device failures in the sarcopenic cohort compared to none in the non-sarcopenic cohort. Conclusions: Sarcopenia was prevalent among AUS patients but did not significantly impact overall device survival. These findings suggest that AUS placement may be feasible to perform in well-selected sarcopenic patients.

Artificial intelligence (AI) promises to revolutionize healthcare. Early identification of disease, appropriate test selection, and automation of repetitive tasks are expected to optimize cost-effective care delivery. However, pragmatic selection and integration of AI algorithms to enable this transformation remain challenging. Healthcare leaders must navigate complex decisions regarding AI deployment, considering factors such as cost of implementation, benefits to patients and providers, and institutional readiness for adoption. A successful strategy needs to align AI adoption with institutional priorities, select appropriate algorithms to be purchased or internally developed, and ensure adequate support and infrastructure. Further, successful deployment requires algorithm validation and workflow integration to ensure efficacy and usability. User-centric design principles and usability testing are critical for AI adoption, ensuring seamless integration into clinical workflows. Once deployed, continuous improvement processes and ongoing algorithm support ensure continuous benefits to the clinical practice. Vigilant planning and execution are necessary to navigate the complexities of AI implementation in the healthcare environment. By applying the framework outlined in this article, institutions can navigate the ever evolving and complex environment of AI in healthcare to maximize the benefits of these innovative technologies.

Background Neurofibromatosis type 2 (NF2)-related schwannomatosis is an autosomal dominant tumor-predisposition syndrome characterized by bilateral vestibular schwannomas (VS). In patients with VS associated with NF2, vascular endothelial growth factor A inhibitor, bevacizumab, is a systemic treatment option. The aim of this study is to retrospectively evaluate NF2 patient responses to bevacizumab on VS growth and symptom progression.Methods This is a retrospective analysis of patients seen at the Mayo Clinic Rochester Multidisciplinary NF2 Clinic.Results Out of 76 patients with NF2 evaluated between 2020 and 2022, we identified 19 that received treatment with bevacizumab. Thirteen of these patients discontinued bevacizumab after median treatment duration of 12.2 months. The remaining 6 patients are currently receiving bevacizumab treatment for a median duration of 9.4 months as of March, 2023. Fifteen patients had evaluable brain MRI data, which demonstrated partial responses in 5 patients, stable disease in 8, and progression in 2. Within 6 months of bevacizumab discontinuation, 5 patients had rebound growth of their VS greater than 20% from their previous tumor volume, while 3 did not. Three patients with rebound growth went on to have surgery or irradiation for VS management.Conclusions Our single-institution experience confirms prior studies that bevacizumab can control progression of VS and symptoms associated with VS growth. However, we note that there is the potential for rapid VS growth following bevacizumab discontinuation, for which we propose heightened surveillance imaging and symptom monitoring for at least 6 months upon stopping anti-VEGF therapy.

OBJECTIVE:To evaluate the performance of an internally developed and previously validated artificial intelligence (AI) algorithm for magnetic resonance (MR)-derived total kidney volume (TKV) in autosomal dominant polycystic kidney disease (ADPKD) when implemented in clinical practice. PATIENTS AND METHODS:The study included adult patients with ADPKD seen by a nephrologist at our institution between November 2019 and January 2021 and undergoing an MR imaging examination as part of standard clinical care. Thirty-three nephrologists ordered MR imaging, requesting AI-based TKV calculation for 170 cases in these 161 unique patients. We tracked implementation and performance of the algorithm over 1 year. A radiologist and a radiology technologist reviewed all cases (N=170) for quality and accuracy. Manual editing of algorithm output occurred at radiology or radiology technologist discretion. Performance was assessed by comparing AI-based and manually edited segmentations via measures of similarity and dissimilarity to ensure expected performance. We analyzed ADPKD severity class assignment of algorithm-derived vs manually edited TKV to assess impact. RESULTS:Clinical implementation was successful. Artificial intelligence algorithm-based segmentation showed high levels of agreement and was noninferior to interobserver variability and other methods for determining TKV. Of manually edited cases (n=84), the AI-algorithm TKV output showed a small mean volume difference of -3.3%. Agreement for disease class between AI-based and manually edited segmentation was high (five cases differed). CONCLUSION:Performance of an AI algorithm in real-life clinical practice can be preserved if there is careful development and validation and if the implementation environment closely matches the development conditions.

You have accessJournal of UrologyCME1 Apr 2023MP65-16 AI DRIVEN ASSESSMENT OF BODY COMPOSITION PARAMETERS IN RADICAL CYSTECTOMY PATIENTS: PREDICTORS OF 90-DAY COMPLICATIONS Anthony Fadel, Vidit Sharma, Matthew K. Tollefson, Daniel J. Blezek, Robert F. Tarrell, Prabin Thapa, Lyndsay D. Viers, Aaron M. Potretzke, Stephen A. Boorjian, Igor Frank, Robert P. Hartman, and Boyd R. Viers Anthony FadelAnthony Fadel More articles by this author , Vidit SharmaVidit Sharma More articles by this author , Matthew K. TollefsonMatthew K. Tollefson More articles by this author , Daniel J. BlezekDaniel J. Blezek More articles by this author , Robert F. TarrellRobert F. Tarrell More articles by this author , Prabin ThapaPrabin Thapa More articles by this author , Lyndsay D. ViersLyndsay D. Viers More articles by this author , Aaron M. PotretzkeAaron M. Potretzke More articles by this author , Stephen A. BoorjianStephen A. Boorjian More articles by this author , Igor FrankIgor Frank More articles by this author , Robert P. HartmanRobert P. Hartman More articles by this author , and Boyd R. ViersBoyd R. Viers More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003323.16AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Sarcopenia is associated with increased mortality after radical cystectomy (RCx). Traditional imaging techniques used to assess sarcopenia are time consuming and labor intensive. Herein we demonstrate the utility of an AI algorithm with deep learning to analyze CT scans and produce body composition parameters in a time-efficient manner. This allows for outcome prediction and correlation of body measures to post-RCx complications. METHODS: Perioperative CT images for 843 RCx patients from 2009-2017 were collected from our institution. An AI algorithm was developed to extract muscle and adipose tissue parameters from 2D axial images at the L3 level. The following areas were segmented: skeletal muscle (SM), visceral adipose tissue (VAT), and subcutaneous adipose tissue (SAT). Skeletal muscle index (SMI) and fat mass index (FMI) were then calculated. All measures were correlated with post-RCx complications using multivariable logistic regression analysis. RESULTS: There was significant variation in pre-operative body composition (Figure 1). An FMI>208 was associated with significantly more wound complications (40% vs 19%, p<.001) while an FMI>260 was associated with more infectious complications (38% vs 21%, p=.003). After adjusting for patient characteristics, these associations of FMI were maintained on multivariable analysis for more infectious (Odds ratio (OR) 1.004, p=.002) and wound (OR 1.006, p<.001) complications. When examining the components of FMI, SAT was independently associated with more wound complications (OR 1.003, p=.006) whereas VAT was independently associated with increased odds of 90-day infectious complications (OR 1.002, p=.011). Similarly, an SMI<42 was associated with major complications (28% vs 17%, p=.002), and on multivariable analysis higher pre-operative SMI was associated with lower odds of major complications (OR 0.972, p=.008). CONCLUSIONS: An AI algorithm was successfully able to segment body composition areas of adipose and skeletal muscle tissues. Sarcopenia assessment using this AI technology is now clinically feasible. Changes in body parameters corresponded with changes in body indices and were predictive of wound, infectious, and major complications. Source of Funding: None. © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e897 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Anthony Fadel More articles by this author Vidit Sharma More articles by this author Matthew K. Tollefson More articles by this author Daniel J. Blezek More articles by this author Robert F. Tarrell More articles by this author Prabin Thapa More articles by this author Lyndsay D. Viers More articles by this author Aaron M. Potretzke More articles by this author Stephen A. Boorjian More articles by this author Igor Frank More articles by this author Robert P. Hartman More articles by this author Boyd R. Viers More articles by this author Expand All Advertisement PDF downloadLoading ...

You have accessJournal of UrologyCME1 Apr 2023MP72-13 AI CHARACTERIZATION OF LONGITUDINAL CHANGES IN BODY COMPOSITION: PREDICTORS OF CHANGE AFTER RADICAL CYSTECTOMY Anthony Fadel, Vidit Sharma, Matthew K. Tollefson, Daniel J. Blezek, Robert F. Tarrell, Prabin Thapa, Lyndsay D. Viers, Aaron M. Potretzke, Stephen A. Boorjian, Igor Frank, Robert P. Hartman, and Boyd R. Viers Anthony FadelAnthony Fadel More articles by this author , Vidit SharmaVidit Sharma More articles by this author , Matthew K. TollefsonMatthew K. Tollefson More articles by this author , Daniel J. BlezekDaniel J. Blezek More articles by this author , Robert F. TarrellRobert F. Tarrell More articles by this author , Prabin ThapaPrabin Thapa More articles by this author , Lyndsay D. ViersLyndsay D. Viers More articles by this author , Aaron M. PotretzkeAaron M. Potretzke More articles by this author , Stephen A. BoorjianStephen A. Boorjian More articles by this author , Igor FrankIgor Frank More articles by this author , Robert P. HartmanRobert P. Hartman More articles by this author , and Boyd R. ViersBoyd R. Viers More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003340.13AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Long term body composition changes after radical cystectomy (RCx) are unknown. Herein, we characterize changes in body composition over one year after RCx and determine predictors of unfavorable change in radiographic nutritional indices. METHODS: CT images for 843 RCx patients from 2009-2017 were collected at baseline, 3 months, and 1 year. Artificial intelligence algorithms extracted two-dimensional axial skeletal muscle and adipose areas at the L3 level. Skeletal muscle index (SMI) and fat mass index (FMI) were calculated, and multivariable logistic regression models were performed to determine factors associated with unfavorable body composition change: absolute top tertile change in FMI and SMI. RESULTS: Muscle and fat body parameters decreased from baseline to 3 months but increased from 3 months to 1 year (Figure 1). However, the distributions of changes in SMI and FMI across these timepoints (Figure 2) show that some patients were able to maintain muscle and fat mass after surgery. In the first 3 months, increasing age and higher BMI were associated with greater loss in FMI (odds ratio (OR) < 1, p<.05); while neoadjuvant chemotherapy was associated with increased gains (OR 1.5, p=.03). After 3 months, robotic surgery was associated with greater gains in FMI (OR 1.8, p=.03). In the first 3 months, on multivariable analysis, men (OR 2.06), increasing BMI (OR 1.08), neoadjuvant chemotherapy (OR 3.98), and any 30-day complications (OR 1.79) were associated with greater loss in SMI from baseline to 3 months (p<.05 for all). After 3 months, older patients had smaller gains in SMI (p<.05), as did patients with bladder cancer recurrences (p<.05). On the contrary, patients with continent diversion were associated with greater gains in SMI (p<.05). CONCLUSIONS: Long-term changes in body composition after RCx are common, and clinically relevant predictors of unfavorable body composition change were identified indicating a catabolic state. Individualized rehabilitation interventions based on these predictors warrants further study. Source of Funding: None © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e1029 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Anthony Fadel More articles by this author Vidit Sharma More articles by this author Matthew K. Tollefson More articles by this author Daniel J. Blezek More articles by this author Robert F. Tarrell More articles by this author Prabin Thapa More articles by this author Lyndsay D. Viers More articles by this author Aaron M. Potretzke More articles by this author Stephen A. Boorjian More articles by this author Igor Frank More articles by this author Robert P. Hartman More articles by this author Boyd R. Viers More articles by this author Expand All Advertisement PDF downloadLoading ...

BACKGROUND: Idiopathic pulmonary fibrosis (IPF) is a progressive, often fatal form of interstitial lung disease (ILD) characterized by the absence of a known cause and usual interstitial pneumonitis (UIP) pattern on chest CT imaging and/or histopathology. Distinguishing UIP/IPF from other ILD subtypes is essential given different treatments and prognosis. Lung biopsy is necessary when noninvasive data are insufficient to render a confident diagnosis. RESEARCH QUESTION: Can we improve noninvasive diagnosis of UIP be improved by predicting ILD histopathology from CT scans by using deep learning? STUDY DESIGN AND METHODS: This study retrospectively identified a cohort of 1,239 patients in a multicenter database with pathologically proven ILD who had chest CT imaging. Each case was assigned a label based on histopathologic diagnosis (UIP or non-UIP). A custom deep learning model was trained to predict class labels from CT images (training set, n = 894) and was evaluated on a 198-patient test set. Separately, two subspecialty-trained radiologists manually labeled each CT scan in the test set according to the 2018 American Thoracic Society IPF guidelines. The performance of the model in predicting histopathologic class was compared against radiologists' performance by using area under the receiveroperating characteristic curve as the primary metric. Deep learning model reproducibility was compared against intra-rater and inter-rater radiologist reproducibility. RESULTS: For the entire cohort, mean patient age was 62 +/- 12 years, and 605 patients were female (49%). Deep learning performance was superior to visual analysis in predicting histopathologic diagnosis (area under the receiver-operating characteristic curve, 0.87 vs 0.80, respectively; P < .05). Deep learning model reproducibility was significantly greater than radiologist inter-rater and intra-rater reproducibility (95% CI for difference in Krippendorff s alpha did not include zero). INTERPRETATION: Deep learning may be superior to visual assessment in predicting UIP/IPF histopathology from CT imaging and may serve as an alternative to invasive lung biopsy.

Ultrasound localization microscopy (ULM) has been proposed to image microvasculature beyond the ultrasound diffraction limit. Although ULM can attain microvascular images with a sub-diffraction resolution, long data acquisition time and processing time are the critical limitations. Deep learning-based ULM (deep-ULM) has been proposed to mitigate these limitations. However, microbubble (MB) localization used in deep-ULMs is currently based on spatial information without the use of temporal information. The highly spatiotemporally coherent MB signals provide a strong feature that can be used to differentiate MB signals from background artifacts. In this study, a deep neural network was employed and trained with spatiotemporal ultrasound datasets to better identify the MB signals by leveraging both the spatial and temporal information of the MB signals. Training, validation and testing datasets were acquired from MB suspension to mimic the realistic intensity-varying and moving MB signals. The performance of the proposed network was first demonstrated in the chicken embryo chorioallantoic membrane dataset with an optical microscopic image as the reference standard. Substantial improvement in spatial resolution was shown for the reconstructed super-resolved images compared with power Doppler images. The full-width-half-maximum (FWHM) of a microvessel was improved from 133 μm to 35 μm, which is smaller than the ultrasound wavelength (73 μm). The proposed method was further tested in an in vivo human liver data. Results showed the reconstructed super-resolved images could resolve a microvessel of nearly 170 μm (FWHM). Adjacent microvessels with a distance of 670 μm, which cannot be resolved with power Doppler imaging, can be well-separated with the proposed method. Improved contrast ratios using the proposed method were shown compared with that of the conventional deep-ULM method. Additionally, the processing time to reconstruct a high-resolution ultrasound frame with an image size of 1024 × 512 pixels was around 16 ms, comparable to state-of-the-art deep-ULMs.

Introduction Sacral tumor resection is known for a high rate of complications. Sarcopenia has been found to be associated with wound complications; however, there is a paucity of data examining the impact of sarcopenia on the outcome of sacral tumor resection. Methods Forty-eight patients (31 primary sarcomas, 17 locally recurrent carcinomas) undergoing sacrectomy were reviewed. Central sarcopenia was assessed by measuring the psoas:lumbar vertebra index (PLVI), with the 50th percentile (0.97) used to determine which patients were high (>0.97) versus low (<0.97). Results Twenty-four (50%) patients had a high PLVI and 24 (50%) had a low PLVI (sarcopenic). There was no difference (p > 0.05) in the demographics of patients with or without sarcopenia. There was no difference in the incidence of postoperative wound complications (odds ratio [OR] = 1.0, p = 1.0) or deep infection (OR = 0.83, p = 1.0). Sarcopenia was not associated with death due to disease (hazard ratio [HR] = 2.04, p = 0.20) or metastatic disease (HR = 2.47, p = 0.17), but was associated with local recurrence (HR = 6.60, p = 0.01). Conclusions Central sarcopenia was not predictive of wound complications or infection following sacral tumor resection. Sarcopenia was, however, an independent risk factor for local tumor recurrence following sacrectomy and should be considered when counseling patients on the outcome of sacrectomy.

Machine learning and artificial intelligence (AI) algorithms hold significant promise for addressing important clinical needs when applied to medical imaging; however, integration of algorithms into a radiology department is challenging. Vended algorithms are integrated into the workflow, successfully, but are typically closed systems and unavailable for site researchers to deploy algorithms. Rather than AI researchers creating one-off solutions, a general, multi-purpose integration system is desired. Here, we present a set of use cases and requirements for a system designed to enable rapid deployment of AI algorithms into the radiologist's workflow. The system uses standards-compliant digital imaging and communications in medicine structured reporting (DICOM SR) to present AI measurements, results, and findings to the radiologist in a clinical context and enables acceptance or rejection of results. The system also implements a feedback mechanism for post-processing technologists to correct results as directed by the radiologist. We demonstrate integration of a body composition algorithm and an algorithm for determining total kidney volume for patients with polycystic kidney disease.

Imaging-based measurements form the basis of surgical decision making in patients with aortic aneurysm. Unfortunately, manual measurement suffer from suboptimal temporal reproducibility, which can lead to delayed or unnecessary intervention. We tested the hypothesis that deep learning could improve upon the temporal reproducibility of CT angiography-derived thoracic aortic measurements in the setting of imperfect ground-truth training data. To this end, we trained a standard deep learning segmentation model from which measurements of aortic volume and diameter could be extracted. First, three blinded cardiothoracic radiologists visually confirmed non-inferiority of deep learning segmentation maps with respect to manual segmentation on a 50-patient hold-out test cohort, demonstrating a slight preference for the deep learning method (p < 1e-5). Next, reproducibility was assessed by evaluating measured change (coefficient of reproducibility and standard deviation) in volume and diameter values extracted from segmentation maps in patients for whom multiple scans were available and whose aortas had been deemed stable over time by visual assessment (n = 57 patients, 206 scans). Deep learning temporal reproducibility was superior for measures of both volume (p < 0.008) and diameter (p < 1e-5) and reproducibility metrics compared favorably with previously reported values of manual inter-rater variability. Our work motivates future efforts to apply deep learning to aortic evaluation.

Recognition of key concepts of structural and functional anatomy of the cerebellum can facilitate image interpretation and clinical correlation. Recently, the human brain mapping literature has increased our understanding of cerebellar anatomy, function, connectivity with the cerebrum, and significance of lesions involving specific areas. Both the common names and numerically based Schmahmann classifications of cerebellar lobules are illustrated. Anatomic patterns, or signs, of key fissures and white matter branching are introduced to facilitate easy recognition of the major anatomic features. Color-coded overlays of cross-sectional imaging are provided for reference of more complex detail. Examples of exquisite detail of structural and functional cerebellar anatomy at 7 T MRI are also depicted. The functions of the cerebellum are manifold with the majority of areas involved with non-motor association function. Key concepts of lesion-symptom mapping which correlates lesion location to clinical manifestation are introduced, emphasizing that lesions in most areas of the cerebellum are associated with predominantly non-motor deficits. Clinical correlation is reinforced with examples of intrinsic pathologic derangement of cerebellar anatomy and altered functional connectivity due to pathology of the cerebral hemisphere. The purpose of this pictorial review is to illustrate basic concepts of these topics in a cross-sectional imaging-based format that can be easily understood and applied by radiologists.