Purpose To evaluate the performance of a deep learning convolutional neural network (CNN) model compared with a traditional natural language processing (NLP) model in extracting pulmonary embolism (PE) findings from thoracic computed tomography (CT) reports from two institutions. Materials and Methods Contrast material-enhanced CT examinations of the chest performed between January 1, 1998, and January 1, 2016, were selected. Annotations by two human radiologists were made for three categories: the presence, chronicity, and location of PE. Classification of performance of a CNN model with an unsupervised learning algorithm for obtaining vector representations of words was compared with the open-source application PeFinder. Sensitivity, specificity, accuracy, and F1 scores for both the CNN model and PeFinder in the internal and external validation sets were determined. Results The CNN model demonstrated an accuracy of 99% and an area under the curve value of 0.97. For internal validation report data, the CNN model had a statistically significant larger F1 score (0.938) than did PeFinder (0.867) when classifying findings as either PE positive or PE negative, but no significant difference in sensitivity, specificity, or accuracy was found. For external validation report data, no statistical difference between the performance of the CNN model and PeFinder was found. Conclusion A deep learning CNN model can classify radiology free-text reports with accuracy equivalent to or beyond that of an existing traditional NLP model. RSNA, 2017 Online supplemental material is available for this article.

This paper explores cutting-edge deep learning methods for information extraction from medical imaging free text reports at a multi-institutional scale and compares them to the state-of-the-art domain-specific rule-based system -PEFinder and traditional machine learning methods -SVM and Adaboost. We proposed two distinct deep learning models -(i) CNN Word -Glove, and (ii) Domain phrase attention-based hierarchical recurrent neural network (DPA-HNN), for synthesizing information on pulmonary emboli (PE) from over 7,370 clin-ical thoracic computed tomography (CT) free-text radiology reports collected from four major healthcare centers. Our proposed DPA-HNN model encodes domain-dependent phrases into an attention mechanism and represents a radiology report through a hierarchical RNN structure composed of word-level, sentence-level and document-level representations. Experimental results suggest that the performance of the deep learning models that are trained on a single institutional dataset, are better than rule-based PEFinder on our multi-institutional test sets. The best F1 score for the presence of PE in an adult patient population was 0.99 (DPA-HNN) and for a pediatrics population was 0.99 (HNN) which shows that the deep learning models being trained on adult data, demonstrated generalizability to pediatrics population with comparable accuracy. Our work suggests feasibility of broader usage of neural network models in automated classification of multi-institutional imaging text reports for a variety of applications including evaluation of imaging utilization, g

No abstract

Sixteen patients with cirrhosis (75% males, 87.5% Caucasians, with mean age of 50.3 þ 10 years) were included. Fifteen patients had decompensated cirrhosis (ascites:93%, VB: 26.7% and HE: 60.0%), while 1 patient had ascites due allograft dysfunction after orthotopic liver transplant (OLT). The mean PSG was reduced from 14.8 ± 5.0 mm Hg pre-TIPS to 5.3 ± 2.7 mm Hg post-TIPS. The mean Model for End-Stage Liver Disease (MELD) score increased from 13.6 ± 4.4 before TIPS to 15.7 ± 4.5 after TIPS. All sixteen patients had abdominal surgeries, one was emergent and 11 were hernia repair. Median time to surgery was 39 days. Median hospital stay was 4 days. The mean expected PO 30-day mortality risk was 27.3 ± 21.6%. Observed 30-day PO mortality rate was found to be 0%. Data was available on 12/16 patients at one year and the observed mortality was 8%. After TIPS, HE was reported in all 16 patients and ascites in 6 patients with 2 requiring revision of TIPS within 30 days after the procedure. One patient, with a PO 30-day mortality risk of 43.6%, died (78 days from the procedure) and 5 patients received liver transplant. Conclusions: Our retrospective review indicates that preoperative TIPS placement in patients with decompensated cirrhosis, especially ascites, can help reduce the PO mortality.

while visualization of varices Z2mm on cross-sectional imaging was confirmed by two independent radiologists. ANOVA, Chi-squared, and Fisher's exact tests were used to assess continuous and categorical data. Results: 24 patients (38.7%) had persistent visible varices (VIS) on follow-up cross-sectional imaging, while 38 patients (61.3%) no longer had visible varices (NON-VIS). Except for age (57.5 yrs vs 50.9 yrs, p ¼ 0.032), preoperative characteristics did not differ between VIS and NON-VIS groups, and mean time to follow-up imaging (VIS 183 days vs NON-VIS 215 days, p ¼ 0.719) was similar. Variceal occlusion method (embolization vs embolization plus sclerotherapy) did not significantly affect the proportion of patients with visible varices (embolization 47.6% vs embolization plus sclerotherapy 36.6%, p ¼ 0.654). In patients with persistent varices, 22 (91.7%) showed esophageal varices, of which 17 (77.3%) protruded intraluminally, while 15 (62.5%) had gastric varices, of which 6 (40%) protruded intraluminally. Mean size of esophageal vs gastric varices did not significantly differ (5.3 mm vs 6.1 mm, p ¼ 0.381). The presence of visible varices did not increase the rate of rebleeding (VIS 0.0%, vs NON-VIS 7.9%, p ¼ 0.16) or rate of repeat variceal occlusion interventions (VIS 0.0% vs NON-VIS 7.9%, p ¼ 0.16). Conclusions: Despite procedures to occlude varices, 38.7% of patients had persistent variceal filling on cross-sectional imaging. The presence of visibly filling varices on cross-sectional imaging after TIPS with embolotherapy did not confer a higher risk for rebleeding. The study is limited by few rebleeding events.

Conclusions: The ALBI score is a significant predictor of mortality after TIPS creation. However, the MELD score remains the superior predictor of mortality after TIPS.

No abstract

No abstract