Diagnostic yield, safety, and outcomes of Head-to-pelvis sudden death CT imaging in post arrest care: The CT FIRST cohort study.

AIM: Our aim was to test whether a head-to-pelvis CT scan improves diagnostic yield and speed to identify causes for out of hospital circulatory arrest (OHCA). METHODS: CT FIRST was a prospective observational pre-/post-cohort study of patients successfully resuscitated from OHCA. Inclusion criteria included unknown cause for arrest, age >18 years, stability to undergo CT, and no known cardiomyopathy or obstructive coronary artery disease. A head-to-pelvis sudden death CT (SDCT) scan within 6 hours of hospital arrival was added to the standard of care for patients resuscitated from OHCA (post-cohort) and compared to standard of care (SOC) alone (pre-cohort). The primary outcome was SDCT diagnostic yield. Secondary outcomes included time to identifying OHCA cause and time-critical diagnoses, SDCT safety, and survival to hospital discharge. RESULTS: Baseline characteristics between the SDCT (N = 104) and the SOC (N = 143) cohorts were similar. CT scans (either head, chest, and/or abdomen) were ordered in 74 (52%) of SOC patients. Adding SDCT scanning identified 92% of causes for arrest compared to 75% (SOC-cohort; p value < 0.001) and reduced the time to diagnosis by 78% (SDCT 3.1 hours, SOC alone 14.1 hours, p < 0.0001). Identification of critical diagnoses was similar between cohorts, but SDCT reduced delayed (>6 hours) identification of critical diagnoses by 81% (p < 0.001). SDCT safety endpoints were similar including acute kidney injury. Patient survival to discharge was similar between cohorts. DISCUSSION: SDCT scanning early after OHCA resuscitation safely improved the efficiency and diagnostic yield for causes of arrest compared to the standard of care alone. CLINICAL TRIALS NUMBER: NCT03111043.

Event-Based Clinical Finding Extraction from Radiology Reports with Pre-trained Language Model.

Radiology reports contain a diverse and rich set of clinical abnormalities documented by radiologists during their interpretation of the images. Comprehensive semantic representations of radiological findings would enable a wide range of secondary use applications to support diagnosis, triage, outcomes prediction, and clinical research. In this paper, we present a new corpus of radiology reports annotated with clinical findings. Our annotation schema captures detailed representations of pathologic findings that are observable on imaging ("lesions") and other types of clinical problems ("medical problems"). The schema used an event-based representation to capture fine-grained details, including assertion, anatomy, characteristics, size, and count. Our gold standard corpus contained a total of 500 annotated computed tomography (CT) reports. We extracted triggers and argument entities using two state-of-the-art deep learning architectures, including BERT. We then predicted the linkages between trigger and argument entities (referred to as argument roles) using a BERT-based relation extraction model. We achieved the best extraction performance using a BERT model pre-trained on 3 million radiology reports from our institution: 90.9-93.4% F1 for finding triggers and 72.0-85.6% F1 for argument roles. To assess model generalizability, we used an external validation set randomly sampled from the MIMIC Chest X-ray (MIMIC-CXR) database. The extraction performance on this validation set was 95.6% for finding triggers and 79.1-89.7% for argument roles, demonstrating that the model generalized well to the cross-institutional data with a different imaging modality. We extracted the finding events from all the radiology reports in the MIMIC-CXR database and provided the extractions to the research community.

First Series Using Ultrasonic Propulsion and Burst Wave Lithotripsy to Treat Ureteral Stones.

PURPOSE: Our goal was to test transcutaneous focused ultrasound in the form of ultrasonic propulsion and burst wave lithotripsy to reposition ureteral stones and facilitate passage in awake subjects. MATERIALS AND METHODS: Adult subjects with a diagnosed proximal or distal ureteral stone were prospectively recruited. Ultrasonic propulsion alone or with burst wave lithotripsy was administered by a handheld transducer to awake, unanesthetized subjects. Efficacy outcomes included stone motion, stone passage, and pain relief. Safety outcome was the reporting of associated anticipated or adverse events. RESULTS: Twenty-nine subjects received either ultrasonic propulsion alone (n = 16) or with burst wave lithotripsy bursts (n = 13), and stone motion was observed in 19 (66%). The stone passed in 18 (86%) of the 21 distal ureteral stone cases with at least 2 weeks follow-up in an average of 3.9±4.9 days post-procedure. Fragmentation was observed in 7 of the burst wave lithotripsy cases. All subjects tolerated the procedure with average pain scores (0-10) dropping from 2.1±2.3 to 1.6±2.0 (P = .03). Anticipated events were limited to hematuria on initial urination post-procedure and mild pain. In total, 7 subjects had associated discomfort with only 2.2% (18 of 820) propulsion bursts. CONCLUSIONS: This study supports the efficacy and safety of using ultrasonic propulsion and burst wave lithotripsy in awake subjects to reposition and break ureteral stones to relieve pain and facilitate passage.

Extracting Radiological Findings With Normalized Anatomical Information Using a Span-Based BERT Relation Extraction Model.

Medical imaging is critical to the diagnosis and treatment of numerous medical problems, including many forms of cancer. Medical imaging reports distill the findings and observations of radiologists, creating an unstructured textual representation of unstructured medical images. Large-scale use of this text-encoded information requires converting the unstructured text to a structured, semantic representation. We explore the extraction and normalization of anatomical information in radiology reports that is associated with radiological findings. We investigate this extraction and normalization task using a span-based relation extraction model that jointly extracts entities and relations using BERT. This work examines the factors that influence extraction and normalization performance, including the body part/organ system, frequency of occurrence, span length, and span diversity. It discusses approaches for improving performance and creating high-quality semantic representations of radiological phenomena.

Impact of Follow-Up Imaging Recommendation Specificity on Adherence.

Radiology reports often contain follow-up imaging recommendations, but failure to comply with them in a timely manner can lead to delayed treatment, poor patient outcomes, complications, and legal liability. Using a dataset containing 2,972,164 exams for over 7 years, in this study we explored the association between recommendation specificity on follow-up rates. Our results suggest that explicitly mentioning the follow-up interval as part of a follow-up imaging recommendation has a significant impact on adherence making these recommendations 3 times more likely (95% CI: 2.95 - 3.05) to be followed-up, while explicit mentioning of the follow-up modality did not have a significant impact. Our findings can be incorporated into routine dictation macros so that the follow-up duration is explicitly mentioned whenever clinically applicable, and/or used as the basis for a quality improvement project focussed on improving adherence to follow-up imaging recommendations.

Correlating the Radiological Assessment of Patient Motion with the Incidence of Repeat Sequences Documented by Log Files.

OBJECTIVE: To correlate a radiological assessment of MR motion artifacts with the incidence of repeated sequences and delays derived from modality log files (MLFs) and investigate the suitability of log files for quantifying the operational impact of patient motion. MATERIALS AND METHODS: An experienced, blinded neuroradiologist retrospectively evaluated one full calendar month of sequentially obtained clinical MR exams of the head and/or brain for the presence of motion artifacts using a previously defined clinical grading scale. MLF data were analyzed to extract the occurrence of repeated sequences during the examinations. Statistical analysis included the determination of 95% confidence intervals for repetition ratios, and Welch's t-test to exclude the hypothesis of equal means for different groups of sequences. RESULTS: A total of 213 examinations were evaluated, comprising 1681 MLF-documented sequences, from which 1580 were archived. Radiological motion assessment scores (0, none to 4, severe) were assigned to each archived sequence. Higher motion scores correlated with a higher MLF-derived repetition probability, reflected by the average motion scores assigned to sequences that would be repeated (group 1, mean=2.5), those that are a repeat (group 2, mean=1.9), and those that are not repeated (group 3, mean=1.1) within an exam. The hypothesis of equal means was rejected with P = 5.9 × 10-5 for groups 1 and 2, P = 9.39 × 10-16 for groups 1 and 3, and P = 1.55 × 10-12 for groups 2 and 3. The repetition probability and associated time loss could be quantified for individual sequence types. The total time loss due to repeat sequence acquisition derived from MLFs was greater than four hours. CONCLUSION: Log file data may help assess patterns of scanner and exam performance and may be useful in identifying pitfalls to diagnostic imaging in a clinical environment, particularly with respect to patient motion.

Prevalence and Patterns of Resuscitation-Associated Injury Detected by Head-to-Pelvis Computed Tomography After Successful Out-of-Hospital Cardiac Arrest Resuscitation.

Background Patients resuscitated from out-of-hospital circulatory arrest (OHCA) frequently have cardiopulmonary resuscitation injuries identifiable by computed tomography, although the prevalence, types of injury, and effects on clinical outcomes are poorly characterized. Methods and Results We assessed the prevalence of resuscitation-associated injuries in a prospective, observational study of a head-to-pelvis sudden-death computed tomography scan within 6 hours of successful OHCA resuscitation. Primary outcomes included total injuries and time-critical injuries (such as organ laceration). Exploratory outcomes were injury associations with mechanical cardiopulmonary resuscitation and survival to discharge. Among 104 patients with OHCA (age 56±15 years, 30% women), 58% had bystander cardiopulmonary resuscitation, and total cardiopulmonary resuscitation time was 15±11 minutes. The prevalence of resuscitation-associated injury was high (81%), including 15 patients (14%) with time-critical findings. Patients with resuscitation injury were older (58±15 versus 46±13 years; P<0.001), but had otherwise similar baseline characteristics and survival compared with those without. Mechanical chest compression systems (27%) had more frequent sternal fractures (36% versus 12%; P=0.009), including displaced fractures (18% versus 1%; P=0.005), but no difference in survival (46% versus 41%; P=0.66). Conclusions In patients resuscitated from OHCA, head-to-pelvis sudden-death computed tomography identified resuscitation injuries in most patients, with nearly 1 in 7 with time-critical complications, and one-half with extensive rib-cage injuries. These data suggest that sudden-death computed tomography may have additional diagnostic utility and treatment implications beyond evaluating causes of OHCA. These important findings need to also be taken in context of the certain fatal outcome without resuscitation efforts. Registration URL: https://www.clinicaltrials.gov; Unique identifier: NCT03111043.

A Method for Reducing Variability Across Dual-Energy CT Manufacturers in Quantification of Low Iodine Content Levels.

BACKGROUND. Clinical use of the dual-energy CT (DECT) iodine quantification technique is hindered by between-platform (i.e., across different manufacturers) variability in iodine concentration (IC) values, particularly at low iodine levels. OBJECTIVE. The purpose of this study was to develop in an anthropomorphic phantom a method for reducing between-platform variability in quantification of low iodine content levels using DECT and to evaluate the method's performance in patients undergoing serial clinical DECT examinations on different platforms. METHODS. An anthropomorphic phantom in three body sizes, incorporating varied lesion types and scanning conditions, was imaged with three distinct DECT implementations from different manufacturers at varying radiation exposures. A cross-platform iodine quantification model for correcting between-platform variability at low iodine content was developed using the phantom data. The model was tested in a retrospective series of 30 patients (20 men, 10 women; median age, 62 years) who each underwent three serial contrast-enhanced DECT examinations of the abdomen and pelvis (90 scans total) for routine oncology surveillance using the same three DECT platforms as in the phantom. Estimated accuracy of phantom IC values was summarized using root-mean-square error (RMSE) relative to known IC. Between-platform variability in patients was summarized using root-mean-square deviation (RMSD). RMSE and RMSD were compared between platform-based IC (ICPB) and cross-platform IC (ICCP). ICPB was normalized to aorta and portal vein. RESULTS. In the phantom study, mean RMSE of ICPB across platforms and other experimental conditions was 0.65 ± 0.18 mg I/mL compared with 0.40 ± 0.08 mg I/mL for ICCP (38% decrease in mean RMSE; p < .05). Intrapatient between-platform variability across serial DECT examinations was higher for ICPB than ICCP (RMSD, 97% vs 88%; p < .001). Between-platform variability was not reduced by normalization of ICPB to aorta (RMSD, 97% vs 101%; p = .12) or portal vein (RMSD, 97% vs 97%; p = .81). CONCLUSION. The developed cross-platform method significantly decreased between-platform variability occurring at low iodine content with platform-based DECT iodine quantification. CLINICAL IMPACT. With further validation, the cross-platform method, which has been implemented as a web-based app, may expand clinical use of DECT iodine quantification, yielding meaningful IC values that reflect tissue biologic viability or treatment response in patients who undergo serial examinations on different platforms.

Management of gastrointestinal bleeding: Society of Abdominal Radiology (SAR) Institutional Survey.

Despite guidelines developed to standardize the diagnosis and management of gastrointestinal (GI) bleeding, significant variability remains in recommendations and practice. The purpose of this survey was to obtain information on practice patterns for the evaluation of overt lower GI bleeding (LGIB) and suspected small bowel bleeding. A 34-question electronic survey was sent to all Society of Abdominal Radiology (SAR) members. Responses were received from 52 unique institutions (40 from the United States). Only 26 (50%) utilize LGIB management guidelines. 32 (62%) use CT angiography (CTA) for initial evaluation in unstable patients. In stable patients with suspected LGIB, CTA is the preferred initial exam at 21 (40%) versus colonoscopy at 24 (46%) institutions. CTA use increases after hours for both unstable (n = 32 vs. 35, 62% vs. 67%) and stable patients (n = 21 vs. 27, 40% vs 52%). CTA is required before conventional angiography for stable (n = 36, 69%) and unstable (n = 15, 29%) patients. 38 (73%) institutions obtain two post-contrast phases for CTA. 49 (94%) institutions perform CT enterography (CTE) for occult small bowel bleeding with capsule endoscopy (n = 26, 50%) and CTE (n = 21, 40%) being the initial test performed. 35 (67%) institutions perform multiphase CTE for occult small bowel bleeding. In summary, stable and unstable patients with overt lower GI are frequently imaged with CTA, while CTE is frequently performed for suspected occult small bowel bleeding.

Gastrointestinal Bleeding at CT Angiography and CT Enterography: Imaging Atlas and Glossary of Terms.

Gastrointestinal (GI) bleeding is a common potentially life-threatening medical condition frequently requiring multidisciplinary collaboration to reach the proper diagnosis and guide management. GI bleeding can be overt (eg, visible hemorrhage such as hematemesis, hematochezia, or melena) or occult (eg, positive fecal occult blood test or iron deficiency anemia). Upper GI bleeding, which originates proximal to the ligament of Treitz, is more common than lower GI bleeding, which arises distal to the ligament of Treitz. Small bowel bleeding accounts for 5-10% of GI bleeding cases commonly manifesting as obscure GI bleeding, where the source remains unknown after complete GI tract endoscopic and imaging evaluation. CT can aid in identifying the location and cause of bleeding and is an important complementary tool to endoscopy, nuclear medicine, and angiography in evaluating patients with GI bleeding. For radiologists, interpreting CT scans in patients with GI bleeding can be challenging owing to the large number of images and the diverse potential causes of bleeding. The purpose of this pictorial review by the Society of Abdominal Radiology GI Bleeding Disease-Focused Panel is to provide a practical resource for radiologists interpreting GI bleeding CT studies that reviews the proper GI bleeding terminology, the most common causes of GI bleeding, key patient history and risk factors, the optimal CT imaging technique, and guidelines for case interpretation and illustrates many common causes of GI bleeding. A CT reporting template is included to help generate radiology reports that can add value to patient care. An invited commentary by Al Hawary is available online. Online supplemental material is available for this article. ©RSNA, 2021.

Early head-to-pelvis computed tomography in out-of-hospital circulatory arrest without obvious etiology.

OBJECTIVES: Patients resuscitated from an out-of-hospital circulatory arrest (OHCA) commonly present without an obvious etiology. We assessed the diagnostic capability and safety of early head-to-pelvis computed tomography (CT) imaging in such patients. METHODS: From November 2015 to February 2018, we enrolled 104 patients resuscitated from OHCA without obvious cause (idiopathic OHCA) to an early sudden-death CT (SDCT) scan protocol within 6 h of hospital arrival. The SDCT protocol included a noncontrast CT head, an electrocardiogram-gated cardiac and thoracic CT angiogram, and a nongated venous-phase abdominopelvic CT angiogram. Patients needing urgent cardiac catheterization or hemodynamically unable to tolerate SDCT were excluded. Cardiac CT analyses were blinded, but other SDCT findings were clinically available. Primary endpoints were the number of OHCA causes identified by SDCT compared to the adjudicated cause and critical diagnoses identified by SDCT, including resuscitation complications. Safety endpoints were acute kidney injury (AKI) and inappropriate treatments based on SDCT findings. Acute coronary syndrome was the presumed etiology if any major coronary artery had a >50% stenosis without another OHCA cause. RESULTS: SDCT scans occurred within 1.9 ± 1.0 h of hospital arrival and identified 39% (41/104) of all OHCA causes and 95% (39/41) of causes potentially identifiable by SDCT. Critical findings were identified by SDCT in 98% (43/44) of patients that included potentially life-threatening resuscitation complications of liver or spleen laceration (n = 6); pneumothorax or thoracic organ laceration (n = 8); and mediastinal, pericardial, or vascular hemorrhage (n = 3). SDCT exclusively identified 13 (13%) OHCA causes that would otherwise not be identified without SDCT imaging. No inappropriate treatments resulted from SDCT findings. AKI was common (28%) but only one (1%) patient required new dialysis. CONCLUSIONS: This observational cohort study suggests that early SDCT scanning is safe, can expedite the diagnosis of potential causes, and can meaningfully change clinical management after idiopathic OHCA.

Automatic Assignment of Radiology Examination Protocols Using Pre-trained Language Models with Knowledge Distillation.

Selecting radiology examination protocol is a repetitive, and time-consuming process. In this paper, we present a deep learning approach to automatically assign protocols to computed tomography examinations, by pre-training a domain-specific BERT model (BERTrad). To handle the high data imbalance across exam protocols, we used a knowledge distillation approach that up-sampled the minority classes through data augmentation. We compared classification performance of the described approach with n-gram models using Support Vector Machine (SVM), Gradient Boosting Machine (GBM), and Random Forest (RF) classifiers, as well as the BERTbase model. SVM, GBM and RF achieved macro-averaged F1 scores of 0.45, 0.45, and 0.6 while BERTbase and BERTrad achieved 0.61 and 0.63. Knowledge distillation boosted performance on the minority classes and achieved an F1 score of 0.66.

Diagnostic accuracy of early computed tomographic coronary angiography to detect coronary artery disease after out-of-hospital circulatory arrest.

AIM: To test the diagnostic accuracy of ECG-gated coronary computed tomography angiography (CCTA) to detect coronary artery disease (CAD) among survivors of out-of-hospital circulatory arrest (OHCA). METHODS: We prospectively studied head-to-pelvis computed tomography (CT) scanning (<6 h from hospital arrival) in OHCA survivors. This sub-study tested the primary outcome of CCTA diagnostic accuracy to identify obstructive CAD (≥50% stenosis) compared to clinically-ordered invasive coronary angiography. Patients were not optimized with beta receptor blockade or nitroglycerin. Secondary analyses included CCTA accuracy for CAD in major coronary arteries, obstructive disease at ≥70% stenosis threshold, and where non-evaluable CCTA segments were considered either obstructive or non-obstructive. RESULTS: Of the 104 enrolled OHCA survivors, 28 (27%) received both CT and invasive angiography in this sub study. All CCTA studies were evaluable although 49/346 (14%) individual coronary segments were unevaluable, primarily due to being too small to evaluate (65%). Patient-level diagnostic accuracy for the ≥50% stenosis threshold was high at 0.93 (95% CI 0.77-0.98) with a specificity of 1.0 (95% CI 0.8-1.0), sensitivity of 0.85 (95%CI 0.58-0.96), negative predictive value of 0.88 (95% CI 0.66-0.97) and positive predictive value of 1.0 (0.74-1.0). When non-evaluable segments were considered obstructive, the sensitivity rose to 0.92 (95% CI 0.67-0.99) with lower specificity of 0.27 (95% CI 0.11-0.52). CONCLUSION: Early CCTA of OHCA survivors has high diagnostic accuracy to detect obstructive coronary artery disease. However, the number of non-diagnostic coronary segments is high suggesting further CCTA refinement is needed, such as the pre-CCTA use of nitroglycerin. CLINICAL TRIAL REGISTRATION: NCT03111043 https://clinicaltrials.gov/ct2/show/record/NCT03111043.

Determining Follow-Up Imaging Study Using Radiology Reports.

Radiology reports often contain follow-up imaging recommendations. Failure to comply with these recommendations in a timely manner can lead to delayed treatment, poor patient outcomes, complications, unnecessary testing, lost revenue, and legal liability. The objective of this study was to develop a scalable approach to automatically identify the completion of a follow-up imaging study recommended by a radiologist in a preceding report. We selected imaging-reports containing 559 follow-up imaging recommendations and all subsequent reports from a multi-hospital academic practice. Three radiologists identified appropriate follow-up examinations among the subsequent reports for the same patient, if any, to establish a ground-truth dataset. We then trained an Extremely Randomized Trees that uses recommendation attributes, study meta-data and text similarity of the radiology reports to determine the most likely follow-up examination for a preceding recommendation. Pairwise inter-annotator F-score ranged from 0.853 to 0.868; the corresponding F-score of the classifier in identifying follow-up exams was 0.807. Our study describes a methodology to automatically determine the most likely follow-up exam after a follow-up imaging recommendation. The accuracy of the algorithm suggests that automated methods can be integrated into a follow-up management application to improve adherence to follow-up imaging recommendations. Radiology administrators could use such a system to monitor follow-up compliance rates and proactively send reminders to primary care providers and/or patients to improve adherence.

Society of abdominal radiology gastrointestinal bleeding disease-focused panel consensus recommendations for CTA technical parameters in the evaluation of acute overt gastrointestinal bleeding.

PURPOSE: To formulate consensus recommendations for CT angiography technical parameters used to evaluate overt gastrointestinal (GI) bleeding. METHODS: An electronic questionnaire consisting of 17 questions was sent to a panel of 16 radiologists with expertise on the imaging of GI bleeding from the Society of Abdominal Radiology GI Bleeding disease-focused panel to obtain consensus agreement on issues related to CTA technical parameters for imaging overt GI bleeding. A multi-round Delphi method of voting was performed to obtain consensus which was defined as ≥ 80% agreement. RESULTS: Consensus agreement was reached in 15/17 (89%) of the questions including the technique for the administration of IV contrast, the number of phases, scan timing, and image reconstruction. CONCLUSIONS: A panel of experts on the imaging of GI bleeding from the Society of Abdominal Radiology was able to reach consensus on the majority of technical parameters used for CTA of overt GI bleeding. These recommendations should improve the quality of patient care by adopting these minimal technical requirements for optimal exam performance and lead to less variation in the performance of these exams which will facilitate collecting and comparing published data from different centers. These recommendations will need revisions as additional scientific data become available.

Clinical and Payer-Based Analysis of Value of Dual-Energy Computed Tomography for Workup of Incidental Abdominal Findings.

OBJECTIVE: To perform a clinical and payer-based analysis of the value of dual-energy computed tomography (DECT) for workup of incidental abdominal findings. METHODS: This was a single-center, retrospectively designed, Health Insurance Portability and Accountability Act-compliant study approved by our institutional review board. Sixty-nine examinations in 69 patients (45 men, 24 women; mean age, 57.7 years) who underwent single-phase postcontrast abdominal DECT studies between January 1, 2011, and December 31, 2017, were included. Two radiologists, blinded to study objective and design, reviewed all cases and identified incidental abdominal findings needing further imaging. All incidental findings were reviewed by 2 other investigators, who determined whether an imaging-based diagnosis could be made using DECT virtual noncontrast images and iodine maps. Additional studies and associated payer-reimbursement amounts avoided by use of DECT were estimated. All imaging costs were estimated based on the US Centers for Medicare & Medicaid Services reimbursement amounts. RESULTS: Thirty-four incidental findings (renal mass, n = 20; adrenal nodule, n = 8; pancreatic cystic lesions, n = 3; others, n = 3) were identified in 19 (27.5%) of 69 patients. Dual-energy computed tomography characterized 27 incidental findings in 15 patients and accounted for cost savings of 15 additional imaging examinations (abdominal magnetic resonance imaging, n = 11; abdominal computed tomography, n = 4). Based on Centers for Medicare & Medicaid Services reimbursement amounts, we estimated that, by abolishing the need for additional imaging use, DECT saved US $84.95 per patient. CONCLUSIONS: Dual-energy computed tomography can provide an imaging-based diagnosis of incidental abdominal findings, otherwise incompletely characterized on routine abdominal computed tomography, in approximately 21% of patients. In select patients, the monetary savings from abolishing additional imaging may reduce payer costs associated with use of DECT.

Environmental Factors Predictive of No-Show Visits in Radiology: Observations of Three Million Outpatient Imaging Visits Over 16 Years.

PURPOSE: To evaluate the impact of environmental and socioeconomic factors on outpatient cancellations and "no-show visits" (NSVs) in radiology. MATERIALS AND METHODS: We conducted a retrospective analysis by collecting environmental factor data related to outpatient radiology visits occurring between 2000 and 2015 at our multihospital academic institution. Appointment attendance records were joined with daily weather observations from the National Oceanic and Atmospheric Administration and estimated median income from the US Census American Community Survey. A multivariate logistic regression model was built to examine relationships between NSV rate and median income, commute distance, maximum daily temperature, and daily snowfall. RESULTS: There were 270,574 (8.0%) cancellations and 87,407 (2.6%) NSVs among 3,379,947 scheduled outpatient radiology appointments and 575,206 unique patients from 2000 to 2015. Overall cancellation rates decreased from 14% to 8%, and NSV rates decreased from 6% to 1% as median income increased from $20,000 to $120,000 per year. In a multivariate model, the odds of NSV decreased 10.7% per $10,000 increase in median income (95% confidence interval [CI]: 10.3%-11.1%) and 2.0% per 10°F increase in maximum daily temperature (95% CI: 1.3%-1.6%). The odds of NSV increased 1.4% per 10-mile increase in commute distance (95% CI: 1.3%-1.6%) and 4.5% per 1-inch increase in daily snowfall (95% CI: 3.6%-5.3%). Commute distance was more strongly associated with NSV for those in the two lower tertiles of income than the highest tertile (P < .001). CONCLUSION: Environmental factors are strongly associated with patients' attendance at scheduled outpatient radiology examinations. Modeling of appointment failure risk based on environmental features can help increase the attendance of outpatient radiology appointments.

Automated Tracking of Follow-Up Imaging Recommendations.

OBJECTIVE. Radiology reports often contain follow-up imaging recommendations. Failure to comply with these recommendations in a timely manner can lead to poor patient outcomes, complications, and legal liability. As such, the primary objective of this research was to determine adherence rates to follow-up recommendations. MATERIALS AND METHODS. Radiology-related examination data, including report text, for examinations performed between June 1, 2015, and July 31, 2017, were extracted from the radiology departments at the University of Washington (UW) and Lahey Hospital and Medical Center (LHMC). The UW dataset contained 923,885 examinations, and the LHMC dataset contained 763,059 examinations. A 1-year period was used for detection of imaging recommendations and up to 14-months for the follow-up examination to be performed. RESULTS. On the basis of an algorithm with 97.9% detection accuracy, the follow-up imaging recommendation rate was 11.4% at UW and 20.9% at LHMC. Excluding mammography examinations, the overall follow-up imaging adherence rate was 51.9% at UW (range, 44.4% for nuclear medicine to 63.0% for MRI) and 52.0% at LHMC (range, 30.1% for fluoroscopy to 63.2% for ultrasound) using a matcher algorithm with 76.5% accuracy. CONCLUSION. This study suggests that follow-up imaging adherence rates vary by modality and between sites. Adherence rates can be influenced by various legitimate factors. Having the capability to identify patients who can benefit from patient engagement initiatives is important to improve overall adherence rates. Monitoring of follow-up adherence rates over time and critical evaluation of variation in recommendation patterns across the practice can inform measures to standardize and help mitigate risk.

Automatic Normalization of Anatomical Phrases in Radiology Reports Using Unsupervised Learning.

In today's radiology workflow, free-text reporting is established as the most common medium to capture, store, and communicate clinical information. Radiologists routinely refer to prior radiology reports of a patient to recall critical information for new diagnosis, which is quite tedious, time consuming, and prone to human error. Automatic structuring of report content is desired to facilitate such inquiry of information. In this work, we propose an unsupervised machine learning approach to automatically structure radiology reports by detecting and normalizing anatomical phrases based on the Systematized Nomenclature of Medicine-Clinical Terms (SNOMED CT) ontology. The proposed approach combines word embedding-based semantic learning with ontology-based concept mapping to derive the desired concept normalization. The word embedding model was trained using a large corpus of unlabeled radiology reports. Fifty-six anatomical labels were extracted from SNOMED CT as class labels of the whole human anatomy. The proposed framework was compared against a number of state-of-the-art supervised and unsupervised approaches. Radiology reports from three different clinical sites were manually labeled for testing. The proposed approach outperformed other techniques yielding an average precision of 82.6%. The proposed framework boosts the coverage and performance of conventional approaches for concept normalization, by applying word embedding techniques in semantic learning, while avoiding the challenge of having access to a large amount of annotated data, which is typically required for training classifiers.