Text Report Analysis to Identify Opportunities for Optimizing Target Selection for Chest Radiograph Artificial Intelligence Models.

Our goal was to analyze radiology report text for chest radiographs (CXRs) to identify imaging findings that have the most impact on report length and complexity. Identifying these imaging findings can highlight opportunities for designing CXR AI systems which increase radiologist efficiency. We retrospectively analyzed text from 210,025 MIMIC-CXR reports and 168,949 reports from our local institution collected from 2019 to 2022. Fifty-nine categories of imaging finding keywords were extracted from reports using natural language processing (NLP), and their impact on report length was assessed using linear regression with and without LASSO regularization. Regression was also used to assess the impact of additional factors contributing to report length, such as the signing radiologist and use of terms of perception. For modeling CXR report word counts with regression, mean coefficient of determination, R2, was 0.469 ± 0.001 for local reports and 0.354 ± 0.002 for MIMIC-CXR when considering only imaging finding keyword features. Mean R2 was significantly less at 0.067 ± 0.001 for local reports and 0.086 ± 0.002 for MIMIC-CXR, when only considering use of terms of perception. For a combined model for the local report data accounting for the signing radiologist, imaging finding keywords, and terms of perception, the mean R2 was 0.570 ± 0.002. With LASSO, highest value coefficients pertained to endotracheal tubes and pleural drains for local data and masses, nodules, and cavitary and cystic lesions for MIMIC-CXR. Natural language processing and regression analysis of radiology report textual data can highlight imaging targets for AI models which offer opportunities to bolster radiologist efficiency.

Perspectives on lung visualization: Three-dimensional anatomical modeling of computed and micro-computed tomographic data in comparative evolutionary morphology and medicine with applications for COVID-19.

The vertebrate respiratory system is challenging to study. The complex relationship between the lungs and adjacent tissues, the vast structural diversity of the respiratory system both within individuals and between taxa, its mobility (or immobility) and distensibility, and the difficulty of quantifying and visualizing functionally important internal negative spaces have all impeded descriptive, functional, and comparative research. As a result, there is a relative paucity of three-dimensional anatomical information on this organ system in all vertebrate groups (including humans) relative to other regions of the body. We present some of the challenges associated with evaluating and visualizing the vertebrate respiratory system using computed and micro-computed tomography and its subsequent digital segmentation. We discuss common mistakes to avoid when imaging deceased and live specimens and various methods for merging manual and threshold-based segmentation approaches to visualize pulmonary tissues across a broad range of vertebrate taxa, with a particular focus on sauropsids (reptiles and birds). We also address some of the recent work in comparative evolutionary morphology and medicine that have used these techniques to visualize respiratory tissues. Finally, we provide a clinical study on COVID-19 in humans in which we apply modeling methods to visualize and quantify pulmonary infection in the lungs of human patients.

Radiomics in Abdominopelvic Solid-Organ Oncologic Imaging: Current Status.

Radiomics is the process of extraction of high-throughput quantitative imaging features from medical images. These features represent noninvasive quantitative biomarkers that go beyond the traditional imaging features visible to the human eye. This article first reviews the steps of the radiomics pipeline, including image acquisition, ROI selection and image segmentation, image preprocessing, feature extraction, feature selection, and model development and application. Current evidence for the application of radiomics in abdominopelvic solid-organ cancers is then reviewed. Applications including diagnosis, subtype determination, treatment response assessment, and outcome prediction are explored within the context of hepatobiliary and pancreatic cancer, renal cell carcinoma, prostate cancer, gynecologic cancer, and adrenal masses. This literature review focuses on the strongest available evidence, including systematic reviews, meta-analyses, and large multicenter studies. Limitations of the available literature are highlighted, including marked heterogeneity in radiomics methodology, frequent use of small sample sizes with high risk of overfitting, and lack of prospective design, external validation, and standardized radiomics workflow. Thus, although studies have laid a foundation that supports continued investigation into radiomics models, stronger evidence is needed before clinical adoption.

A Framework for the Virtual Medical Interview Process: Considerations for the Applicant and the Interviewer.

Background: Videoconferencing platforms are being used for the purposes of interviewing in academic medicine because of the coronavirus disease 2019 pandemic. We present considerations applicable to interviewers and interviewees in the virtual space, with a focus on medical school and residency applicants. Methods: We reviewed the literature regarding the virtual interview process for medical school and residency by searching PubMed using the following keywords and terms: "interview," "academic medicine," "medical school application," "residency application," "virtual interviews," and "videoconferencing." Our search identified 701 results, from which we selected 36 articles for review. Results: The garnered information focuses on strategies for optimizing the virtual interview process from the standpoint of both the interviewer and the interviewee. We discuss the advantages and disadvantages of the virtual interview process and present recommendations. Conclusion: While the future of the interview process for medical school and residency is uncertain, virtual interviewing is a common and growing practice that will continue to be at least part of the medical interview process for years to come. Interviewers and interviewees should prepare to adapt to the evolving changes in the process.

Artificial Intelligence in Imaging of Chronic Liver Diseases: Current Update and Future Perspectives.

Here we review artificial intelligence (AI) models which aim to assess various aspects of chronic liver disease. Despite the clinical importance of hepatocellular carcinoma in the setting of chronic liver disease, we focus this review on AI models which are not lesion-specific and instead review models developed for liver parenchyma segmentation, evaluation of portal circulation, assessment of hepatic fibrosis, and identification of hepatic steatosis. Optimization of these models offers the opportunity to potentially reduce the need for invasive procedures such as catheterization to measure hepatic venous pressure gradient or biopsy to assess fibrosis and steatosis. We compare the performance of these AI models amongst themselves as well as to radiomics approaches and alternate modality assessments. We conclude that these models show promising performance and merit larger-scale evaluation. We review artificial intelligence models that aim to assess various aspects of chronic liver disease aside from hepatocellular carcinoma. We focus this review on models for liver parenchyma segmentation, evaluation of portal circulation, assessment of hepatic fibrosis, and identification of hepatic steatosis. We conclude that these models show promising performance and merit a larger scale evaluation.

LI-RADS treatment response assessment of combination locoregional therapy for HCC.

HCC incidence continues to increase worldwide and is most frequently discovered at an advanced stage when limited curative options are available. Combination locoregional therapies have emerged to improve patient survival and quality of life or downstage patients to curative options. The increasing options for locoregional therapy combinations require an understanding of the expected post-treatment imaging appearance in order to assess treatment response. This review aims to describe the synergy between TACE combined with thermal ablation and TACE combined with SBRT. We will also illustrate expected imaging findings that determine treatment efficacy based on the mechanism of tissue injury using the LI-RADS Treatment Response Algorithm.

Assessing locoregional treatment response to Hepatocellular Carcinoma: comparison of hepatobiliary contrast agents to extracellular contrast agents.

Cross-sectional imaging with contrast-enhanced magnetic resonance imaging (MRI) is routinely performed in patients with hepatocellular carcinoma (HCC) to assess tumor response to locoregional therapy (LRT). Current response assessment algorithms, such as the Liver Imaging Reporting and Data System (LI-RADS) treatment response algorithm (TRA), allow assessment using conventional gadolinium-based extracellular contrast agents (ECA) for accurate tumor response assessment following LRT. MRI with hepatobiliary agents (HBA) allows an acquisition of hepatobiliary phase (HBP), which is proven to increase sensitivity for detection of observations in at-risk patients, particularly for findings < 2 cm. The use of HBA is not yet incorporated into the TRA; however, it is increasingly used in clinical practice. Few published studies have evaluated the performance of LI-RADS TRA by applying ancillary features related to HBP that has resulted in category adjustment, enabling more sensitive and unequivocal diagnosis. This may help timely management of viable cases, without a significant loss of specificity in comparison with the ECA-based LI-RADS TRA assessment. In this review, we will describe and compare the imaging appearance of treated HCC on MRI using extracellular and hepatobiliary contrast agents and discuss emerging evidence and pitfalls in the assessment of tumor response following LRT with HBA.

Artificial intelligence in assessment of hepatocellular carcinoma treatment response.

Artificial Intelligence (AI) continues to shape the practice of radiology, with imaging of hepatocellular carcinoma (HCC) being of no exception. This article prepared by members of the LI-RADS Treatment Response (TR LI-RADS) work group and associates, presents recent trends in the utility of AI applications for the volumetric evaluation and assessment of HCC treatment response. Various topics including radiomics, prognostic imaging findings, and locoregional therapy (LRT) specific issues will be discussed in the framework of HCC treatment response classification systems with focus on the Liver Reporting and Data System treatment response algorithm (LI-RADS TRA).

Analysis of Potential for User Errors in Mobile Deployment of Radiology Deep Learning for Cardiac Rhythm Device Detection.

We examine how convolutional neural networks (CNNs) for cardiac rhythm device detection can exhibit failures in performance under suboptimal deployment scenarios and examine how medically adversarial image presentation can further impair neural network performance. We validated the publicly available Pacemaker-ID web server and mobile app on 43 local hospital emergency department (ED) cases of patients presenting with a cardiac rhythm device on anterior-posterior (AP) chest radiograph and assessed performance using Cohen's kappa coefficient for inter-rater reliability. To illustrate adversarial performance concerns, we then produced example CNN models using the 65,379 patient MIMIC-CXR chest radiograph retrospective database and evaluated performance with area under the receiver operating characteristic (AUROC). In retrospective review of 43 patients with cardiac rhythm devices on AP chest radiographs during our study period (January 1, 2020 to March 1, 2020), 74.4% (32/43) had device manufacturer information readily available within the electronic medical record. A total of 25.6% of patients (11/43) did not have this information documented in the patient chart and could ostensibly benefit from CNN-based identification of device manufacturer. For patients with known device manufacturer, the Pacemaker-ID prediction was accurate in 87.5% of cases (28/32). Mobile app accuracy varied from 62.5 to 93.75% depending on image capture settings and presentation. Cohen's kappa coefficient varied from 0.448 to 0.897 depending on mobile image capture conditions. For our additional analysis of medically adversarial performance failures with a DenseNet121 trained on MIMIC-CXR images, we showed that an AUROC of 0.9807 ± 0.0051 could be achieved on an example testing dataset while masking a 30% false positive rate in identification of cardiac rhythm devices versus clinically distinct entities such as vagal nerve stimulators. Despite the promise of CNN approaches for cardiac rhythm device analysis on chest radiographs, further study is warranted to assess potential for errors driven by user misuse when deploying these models to mobile devices as well as for cases when performance can be impaired by the presence of other support apparatuses.

Online conferencing software in radiology: Recent trends and utility.

Videoconferencing platforms have recently gained wide attention due to the COVID-19 pandemic, both within and outside of the medical community. This article reviews various applications of online meeting technology to the radiologic community, not only in response to the recent pandemic but also thereafter. Various platform features are outlined and discussed, specifically with respect to collaboration, training, and patient care. Platforms reviewed are GoToMeeting, Microsoft Teams, Skype, WebEx, and Zoom.

Diagnosis in a snap: a pilot study using Snapchat in radiologic didactics.

PURPOSE: To evaluate Snapchat, an image-based social media platform, as a tool for emergency radiologic didactics comparing image interpretation on mobile devices with conventional analysis on a classroom screen. MATERIALS AND METHODS: Seven radiology residents (4 juniors, 3 seniors;4 males, 3 females; 28.4 years old, ± 1.7 years) were shown 5 emergent radiologic cases using Snapchat and 5 cases of similar content and duration on a classroom projector over 4 weeks. All images depicted diagnoses requiring immediate communication to ordering physicians. Performance was scored 0-2 (0 = complete miss, 1 = major finding, but missed the diagnosis, 2 = correct diagnosis) by two attending radiologists in consensus. RESULTS: All residents performed better on Snapchat each week. In weeks 1-4, juniors scored 21/40 (52.5%), 23/40 (57.5%), 19/40 (47.5%), and 18/40 (45%) points using Snapchat compared with 13/40 (32.5%), 23/40 (57.5%), 14/40 (35%), and 13/40 (32.5%), respectively, each week by projector, while seniors scored 19/30 (63.3%), 21/30 (70%), 27/30 (90%), and 21/30 (70%) on Snapchat versus 16/30 (53.3%), 19/30 (63.3%), 20/30 (66.7%), and 20/30 (66.7%) on projector. Four-week totals showed juniors scoring 81/160 (50.6%) on Snapchat and 63/160 (39.4%) by projector compared with seniors scoring 88/120 (73.3%) and 75/120 (62.5%), respectively. Performance on Snapchat was statistically, significantly better than via projector during weeks 1 and 3 (p values 0.0019 and 0.0031). CONCLUSION: Radiology residents interpreting emergency cases via Snapchat showed higher accuracy compared with using a traditional classroom screen. This pilot study suggests that Snapchat may have a role in the digital radiologic classroom's evolution.

Estimation of age in unidentified patients via chest radiography using convolutional neural network regression.

PURPOSE: Patient age has important clinical utility for refining a differential diagnosis in radiology. Here, we evaluate the potential for convolutional neural network models to predict patient age based on anterior-posterior chest radiographs for instances where patients may present for emergency services without the ability to provide this identifying information. METHODS: We used the CheXpert dataset of 224,316 chest radiographs from 65,240 patients to train CNN regression models with ResNet50 and DenseNet121 architectures for prediction of patient age based on anterior-posterior (AP) view chest radiographs. We evaluate these models on both the CheXpert validation dataset and a local hospital case in which a patient initially presented for emergency services intubated and without identification. RESULTS: Mean absolute error (MAE) for our ResNet50 model on the CheXpert dataset is 4.94 years for predicting patient age based on AP chest radiographs. MAE for our DenseNet121 model is 4.69 years. Both models have a correlation coefficient between true patient ages and predicted ages of 0.944. Wilcoxon rank-sum comparison between the two model architectures shows no significant difference (p = 0.33), but both show improvement over a baseline demographic-driven estimation (p < 0.001). CONCLUSIONS: For circumstances in which patients present for healthcare services without readily accessible identification such as in the setting trauma or altered mental status, CNN regression models for age prediction have potential clinical utility for refining estimates related to this missing patient information.

Follow-up Recommendation Rates Associated With Spectral Detector Dual-Energy CT of the Abdomen and Pelvis: A Retrospective Comparison to Single-Energy CT.

BACKGROUND: Dual-energy CT image sets have many applications in abdominopelvic imaging but no demonstrated clinical effect. PURPOSE: To determine the effect of dual-energy CT iodine maps on abdominopelvic imaging follow-up recommendation rates. MATERIALS AND METHODS: Retrospective study of abdominopelvic CTs acquired from April 2017 through June 2018. CT reports were analyzed for radiologic follow-up recommendation and follow-up recommendation reason. Follow-up MRI reports were analyzed for benign or nonbenign diagnosis. CT scans with iodine maps (CTIMs) and conventional CT scans (CCTs) subgroups were compared using χ2 testing. RESULTS: In all, 3,221 abdominopelvic CT scans of 2,401 patients (1,326 men, 1,075 women, mean age 54.1 years) were analyzed; 1,423 were CTIMs and 1,798 were CCTs. Follow-up recommendation rates were not significantly different for CTIMs and CCTs (19.5% and 21.4%, respectively, P = .19). Follow-up recommendations because of incomplete diagnosis were significantly lower in CTIMs (9.1%) than in CCTs (11.9%, P = .01). Follow-up recommendations for MRI and PET/CT were significantly lower in CTIMs (9.6%) than CCTs (13.0%, P = .003). Follow-up MRI outcomes (n = 111) were not different between CTIMs (61.2% benign) and CCTs (59.6%, P = .87). CONCLUSION: Dual-energy CT iodine maps are associated with decreased follow-up examinations because of incomplete diagnosis and decreased recommendations for follow-up MRI, suggesting that abdominopelvic iodine maps may benefit patient care and decrease institutional cost.

The Effect of Image Resolution on Deep Learning in Radiography.

PURPOSE: To examine variations of convolutional neural network (CNN) performance for multiple chest radiograph diagnoses and image resolutions. MATERIALS AND METHODS: This retrospective study examined CNN performance using the publicly available National Institutes of Health chest radiograph dataset comprising 112 120 chest radiographic images from 30 805 patients. The network architectures examined included ResNet34 and DenseNet121. Image resolutions ranging from 32 × 32 to 600 × 600 pixels were investigated. Network training paradigms used 80% of samples for training and 20% for validation. CNN performance was evaluated based on area under the receiver operating characteristic curve (AUC) and label accuracy. Binary output networks were trained separately for each label or diagnosis under consideration. RESULTS: Maximum AUCs were achieved at image resolutions between 256 × 256 and 448 × 448 pixels for binary decision networks targeting emphysema, cardiomegaly, hernias, edema, effusions, atelectasis, masses, and nodules. When comparing performance between networks that utilize lower resolution (64 × 64 pixels) versus higher (320 × 320 pixels) resolution inputs, emphysema, cardiomegaly, hernia, and pulmonary nodule detection had the highest fractional improvements in AUC at higher image resolutions. Specifically, pulmonary nodule detection had an AUC performance ratio of 80.7% ± 1.5 (standard deviation) (0.689 of 0.854) whereas thoracic mass detection had an AUC ratio of 86.7% ± 1.2 (0.767 of 0.886) for these image resolutions. CONCLUSION: Increasing image resolution for CNN training often has a trade-off with the maximum possible batch size, yet optimal selection of image resolution has the potential for further increasing neural network performance for various radiology-based machine learning tasks. Furthermore, identifying diagnosis-specific tasks that require relatively higher image resolution can potentially provide insight into the relative difficulty of identifying different radiology findings. Supplemental material is available for this article. © RSNA, 2020See also the commentary by Lakhani in this issue.

Feedforward Thalamocortical Connectivity Preserves Stimulus Timing Information in Sensory Pathways.

Reliable timing of cortical spikes in response to visual events is crucial in representing visual inputs to the brain. Spikes in the primary visual cortex (V1) need to occur at the same time within a repeated visual stimulus. Two classical mechanisms are employed by the cortex to enhance reliable timing. First, cortical neurons respond reliably to a restricted set of stimuli through their preference for certain patterns of membrane potential due to their intrinsic properties. Second, intracortical networking of excitatory and inhibitory neurons induces lateral inhibition that, through the timing and strength of IPSCs and EPSCs, produces sparse and reliably timed cortical neuron spike trains to be transmitted downstream. Here, we describe a third mechanism that, through preferential thalamocortical synaptic connectivity, enhances the trial-to-trial timing precision of cortical spikes in the presence of spike train variability within each trial that is introduced between LGN neurons in the retino-thalamic pathway. Applying experimentally recorded LGN spike trains from the anesthetized cat to a detailed model of a spiny stellate V1 neuron, we found that output spike timing precision improved with increasing numbers of convergent LGN inputs. The improvement was consistent with the predicted proportionality of [Formula: see text] for n LGN source neurons. We also found connectivity configurations that maximize reliability and that generate V1 cell output spike trains quantitatively similar to the experimental recordings. Our findings suggest a general principle, namely intra-trial variability among converging inputs, that increases stimulus response precision and is widely applicable to synaptically connected spiking neurons.SIGNIFICANCE STATEMENT The early visual pathway of the cat is favorable for studying the effects of trial-to-trial variability of synaptic inputs and intra-trial variability of thalamocortical connectivity on information transmission into the visual cortex. We have used a detailed model to show that there are preferred combinations of the number of thalamic afferents and the number of synapses per afferent that maximize the output reliability and spike-timing precision of cortical neurons. This provides additional insights into how synchrony in thalamic spike trains can reduce trial-to-trial variability to produce highly reliable reporting of sensory events to the cortex. The same principles may apply to other converging pathways where temporally jittered spike trains can reliably drive the downstream neuron and improve temporal precision.

Intrinsic bursting of AII amacrine cells underlies oscillations in the rd1 mouse retina.

In many forms of retinal degeneration, photoreceptors die but inner retinal circuits remain intact. In the rd1 mouse, an established model for blinding retinal diseases, spontaneous activity in the coupled network of AII amacrine and ON cone bipolar cells leads to rhythmic bursting of ganglion cells. Since such activity could impair retinal and/or cortical responses to restored photoreceptor function, understanding its nature is important for developing treatments of retinal pathologies. Here we analyzed a compartmental model of the wild-type mouse AII amacrine cell to predict that the cell's intrinsic membrane properties, specifically, interacting fast Na and slow, M-type K conductances, would allow its membrane potential to oscillate when light-evoked excitatory synaptic inputs were withdrawn following photoreceptor degeneration. We tested and confirmed this hypothesis experimentally by recording from AIIs in a slice preparation of rd1 retina. Additionally, recordings from ganglion cells in a whole mount preparation of rd1 retina demonstrated that activity in AIIs was propagated unchanged to elicit bursts of action potentials in ganglion cells. We conclude that oscillations are not an emergent property of a degenerated retinal network. Rather, they arise largely from the intrinsic properties of a single retinal interneuron, the AII amacrine cell.