Feasibility of Natural Language Processing–Assisted Auditing of Critical Findings in Chest Radiology

ObjectiveTime-sensitive communication of critical imaging findings like pneumothorax or pulmonary embolism to referring physicians is essential for patient safety. The definitive communication is the radiology free-text report. Quality assurance initiatives require that institutions audit these communications, a time-intensive manual task. We propose using a rule-based natural language processing system to improve the process for auditing critical findings communications.MethodsWe present a pilot assessment of the feasibility of using an automated critical finding identification system to assist quality assurance teams’ evaluation of critical findings communication compliance. Our assessment is based on chest imaging reports. Critical findings are identified in radiology reports using pyConTextNLP, an open source Python implementation of the ConText algorithm.ResultsIn our test set, there were 75 reports with critical findings and 591 reports without critical findings. pyConTextNLP correctly identified 69 of the positive cases with 8 false-positives for a sensitivity of 0.92 and a specificity of 0.99.DiscussionNatural language processing can provide valuable assistance to auditing critical findings communications.     

Radiologists’ Experience With Patient Interactions in the Era of Open Access of Patients to Radiology Reports

PurposeThe aim of this study was to evaluate radiologists’ experiences with patient interactions in the era of open access of patients to radiology reports.MethodsThis prospective, nonrandom survey of staff and trainee radiologists (n = 128) at a single large academic institution was performed with approval from the institutional review board with a waiver of the requirement to obtain informed consent. A multiple-choice questionnaire with optional free-text comments was constructed with an online secure platform (REDCap) and distributed via departmental e-mail between June 1 and July 31, 2016. Participation in the survey was voluntary and anonymous, and responses were collected and aggregated via REDCap. Statistical analysis of categorical responses was performed with the χ² test, with statistical significance defined as P < .05.ResultsAlmost three-quarters of surveys (73.4% [94 of 128]) were completed. Staff radiologists represented 54.3% of survey respondents (51 of 94) and trainees 45.7% (43 of 94). Most respondents (78.7% [74 of 94]) found interactions with patients to be a satisfying experience. More than half of radiologists (54.3% [51 of 94]) desired more opportunities for patient interaction, with no significant difference in the proportion of staff and trainee radiologists who desired more patient interaction (56.9% [29 of 51] versus 51.2% [22 of 43], P = .58). Staff radiologists who specialized in vascular and interventional radiology and mammography were significantly more likely to desire more patient interaction compared with other specialists (77.8% [14 of 18] versus 45.5% [15 of 33], P = .03). Only 4.2% of radiologists (4 of 94) found patient interactions to be detrimental to normal workflow, with 19.1% of radiologists (18 of 94) reporting having to spend more than 15 min per patient interaction.ConclusionsMost academic staff and trainee radiologists would like to have more opportunities for patient interaction and consider patient interaction rarely detrimental to workflow.     

Deep Learning for Natural Language Processing in Radiology—Fundamentals and a Systematic Review

PurposeNatural language processing (NLP) enables conversion of free text into structured data. Recent innovations in deep learning technology provide improved NLP performance. We aimed to survey deep learning NLP fundamentals and review radiology-related research.MethodsThis systematic review was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. We searched for deep learning NLP radiology studies published up to September 2019. MEDLINE, Scopus, and Google Scholar were used as search databases.ResultsTen relevant studies published between 2018 and 2019 were identified. Deep learning models applied for NLP in radiology are convolutional neural networks, recurrent neural networks, long short-term memory networks, and attention networks. Deep learning NLP applications in radiology include flagging of diagnoses such as pulmonary embolisms and fractures, labeling follow-up recommendations, and automatic selection of imaging protocols. Deep learning NLP models perform as well as or better than traditional NLP models.ConclusionResearch and use of deep learning NLP in radiology is increasing. Acquaintance with this technology can help prepare radiologists for the coming changes in their field.     

Cone Beam CT–Guided Chemoembolization of Probable Hepatocellular Carcinomas Smaller than 1 cm in Patients at High Risk of Hepatocellular Carcinoma

Purpose

To evaluate the effectiveness and safety of cone-beam computed tomography (CT)–guided chemoembolization for probable hepatocellular carcinomas (HCCs) smaller than 1 cm in patients at high risk for HCC.

Increasing Utilization of Chest Imaging in US Emergency Departments From 1994 to 2015

PurposeThe aim of this study was to assess national and state-specific changes in emergency department (ED) chest imaging utilization from 1994 to 2015.MethodsUsing aggregate 100% Medicare Physician/Supplier Procedure Summary Master Files for 1994 to 2015, the annual frequency of chest imaging in Medicare Part B beneficiaries in the ED setting was identified, and utilization was normalized to annual Medicare enrollment as well as annual ED visits. Using individual Medicare beneficiary 5% research-identifiable files, similar determinations were performed for each state.ResultsBetween 1994 and 2015, per 1,000 beneficiaries, ED utilization of chest radiography and CT increased by 173% (compound annual growth rate [CAGR] 4.9%) and 5,941.8% (CAGR 21.6%). Per 1,000 ED visits, utilization increased by 81% (CAGR 2.9%) and 3,915.4% (CAGR 19.2%), respectively. Across states, utilization was highly variable, with 2015 radiography utilization per 1,000 ED visits ranging from 82 (Wyoming) to 731 (Hawaii) and CT utilization ranging from 18 (Wyoming) to 76 (Hawaii). Between 2004 and 2015, most states demonstrated increases in the utilization of both radiography (maximal increase of CAGR 11.0% in Vermont) and CT (maximal increase of CAGR 21.0% in Maine). Nonetheless, utilization of radiography declined in four states and utilization of CT in a single state.ConclusionsOver the past two decades, ED utilization of chest imaging has increased. This was related not only to an increasing frequency of ED visits but also to increasing utilization per ED visit. Across states, utilization is highly variable, but with radiography and CT both increasing, the use of CT seems additive to, rather than replacing, radiography.     

Damage to Liver Function after TACE of Anticancer Drugs in Hepatocellular Carcinoma： Evaluation of Two Kinds of Anticancer Drugs

Introduction It is well known that transcatheter hepatic arterial chem oem bolization ( TA CE ) can restrict the growth of liver tum ors , but it can also cause inevitable dam age of liver function. A lthough the im pairm ent of liver function should be r…     

Pain Analysis in Patients with Hepatocellular Carcinoma: Irreversible Electroporation versus Radiofrequency Ablation—Initial Observations

Purpose

To retrospectively compare the postprocedure pain of hepatocellular carcinoma treated with irreversible electroporation (IRE) with radiofrequency ablation (RFA).

Methods

This Health Insurance Portability and Accountability Act–compliant, institutional review board–approved study compared postprocedure pain in 21 patients (15 men, six women; mean age 61.5 years) who underwent IRE of 29 intrahepatic lesions (mean size 2.20 cm) in 28 IRE sessions with 22 patients (16 men, six women; mean age 60.2 years) who underwent RFA of 27 lesions (mean size 3.38 cm) in 25 RFA sessions. Pain was determined by patient-disclosed scores with an 11-point numerical rating scale and 24 h cumulative hydromorphone use from patient-controlled analgesia pump. Complications were noted. Statistical significance was evaluated by Fisher’s exact test, the Chi-square test, and Student’s t test.

Results

There was no significant difference in the cumulative hydromorphone dose (1.54 mg (IRE) vs. 1.24 mg (RFA); P = 0.52) and in the mean pain score (1.96 (IRE) vs. 2.25 (RFA); P = 0.70). In nine (32.14 %) of 28 IRE sessions and 11 (44.0 %) of 25 RFA sessions, patients reported no pain. Complications occurred in three (10.7 %) of 28 IRE treatments and included pneumothorax (n = 1), pleural effusion (n = 1), and bleeding in the form of hemothorax (n = 1); one (4 %) of 25 RFA treatments included burn.

Conclusion

IRE is comparable to RFA in the amount of pain that patients experience and the amount of pain medication self-administered. Both modalities were well tolerated by patients. Prospective, randomized trials are necessary to further evaluate these findings.     

Hepatocellular Carcinoma Responding to Superselective Transarterial Chemoembolization: An Issue of Nodule Dimension?

PurposeTo evaluate the per-nodule efficacy of superselective transarterial chemoembolization of hepatocellular carcinomas (HCCs).Materials and MethodsFrom 2006–2009, 271 cirrhotic patients with 635 nodules underwent a first superselective transarterial chemoembolization, repeated “on demand” after local recurrences (LR) or partial responses (PR). Complete response (CR), time to nodule progression (TTnP), and local recurrence rate (LRR), according to three size classes (≤2 cm, 2.1–5 cm, and>5 cm) were evaluated.ResultsAfter the first superselective transarterial chemoembolization, the CR was 64%, sustained over time in 77%, higher in small (68%) and intermediate-size (64%) nodules than in large nodules (25%; P<.001). The LRR was 23%:20% in small, 27% in intermediate, and 67% in large HCCs (P<.05). The median TTnP of large HCCs was 4 months versus 7–9 months for small and intermediate HCCs. The second superselective transarterial chemoembolization achieved a higher CR (63% in LR, 52% in PR) than the third superselective transarterial chemoembolization (32%). Median TTnP after the second superselective transarterial chemoembolization for LR and PR (8 months and 6 months) was longer than after the third superselective transarterial chemoembolization (3.5 months). Nodules≤5 cm had a CR after the first superselective transarterial chemoembolization (66%) and the second superselective transarterial chemoembolization for LR (64%) or PR (55%) higher than after the third superselective transarterial chemoembolization (40%); nodules>5 cm had a CR of 25% after the first superselective transarterial chemoembolization, LR of 50% and PR of 25%, and after the second and third superselective transarterial chemoembolizations, PR of 0%.ConclusionsEffectiveness of superselective transarterial chemoembolization has a clear cutoff above and below 5-cm nodules, with better results in smaller nodules. In HCCs≤5 cm, the efficacy of the first and second superselective transarterial chemoembolizations performed for LR was higher than the second superselective transarterial chemoembolization for PR and the third superselective transarterial chemoembolization. For HCCs>5 cm, retreatment of PR is of little value, and the third cycle is ineffective.     

Balloon-Occluded Transarterial Chemoembolization for the Treatment of Hepatocellular Carcinoma: A Single-Center US Preliminary Experience

This report describes a single-center experience with balloon-occluded transarterial chemoembolization for liver-directed therapy. A total of 26 patients (11 male, 4 female; mean age, 65 y ± 7) with 28 tumors (mean diameter, 2.7 cm; range, 1.1–5.9 cm) were treated. Technical success rate was 100% (28 of 28 cases), with 1 minor complication of left portal vein thrombosis and small liver infarct. Of the 15 tumors analyzed for response, 60% (9 of 15) exhibited complete response, 33.3% (5 of 15) exhibited partial response, and 6.6% (1 of 15) had stable disease on follow-up. Eight patients exhibited overall progression with a new hepatic lesion and a median time to progression of 7.9 months (range, 5–11 mo).