Evaluating GPT-V4 (GPT-4 with Vision) on Detection of Radiologic Findings on Chest Radiographs.

Background Generating radiologic findings from chest radiographs is pivotal in medical image analysis. The emergence of OpenAI's generative pretrained transformer, GPT-4 with vision (GPT-4V), has opened new perspectives on the potential for automated image-text pair generation. However, the application of GPT-4V to real-world chest radiography is yet to be thoroughly examined. Purpose To investigate the capability of GPT-4V to generate radiologic findings from real-world chest radiographs. Materials and Methods In this retrospective study, 100 chest radiographs with free-text radiology reports were annotated by a cohort of radiologists, two attending physicians and three residents, to establish a reference standard. Of 100 chest radiographs, 50 were randomly selected from the National Institutes of Health (NIH) chest radiographic data set, and 50 were randomly selected from the Medical Imaging and Data Resource Center (MIDRC). The performance of GPT-4V at detecting imaging findings from each chest radiograph was assessed in the zero-shot setting (where it operates without prior examples) and few-shot setting (where it operates with two examples). Its outcomes were compared with the reference standard with regards to clinical conditions and their corresponding codes in the International Statistical Classification of Diseases, Tenth Revision (ICD-10), including the anatomic location (hereafter, laterality). Results In the zero-shot setting, in the task of detecting ICD-10 codes alone, GPT-4V attained an average positive predictive value (PPV) of 12.3%, average true-positive rate (TPR) of 5.8%, and average F1 score of 7.3% on the NIH data set, and an average PPV of 25.0%, average TPR of 16.8%, and average F1 score of 18.2% on the MIDRC data set. When both the ICD-10 codes and their corresponding laterality were considered, GPT-4V produced an average PPV of 7.8%, average TPR of 3.5%, and average F1 score of 4.5% on the NIH data set, and an average PPV of 10.9%, average TPR of 4.9%, and average F1 score of 6.4% on the MIDRC data set. With few-shot learning, GPT-4V showed improved performance on both data sets. When contrasting zero-shot and few-shot learning, there were improved average TPRs and F1 scores in the few-shot setting, but there was not a substantial increase in the average PPV. Conclusion Although GPT-4V has shown promise in understanding natural images, it had limited effectiveness in interpreting real-world chest radiographs. © RSNA, 2024 Supplemental material is available for this article.

Three-dimensional volumetric measurements in defining endoscope-guided giant adenoma surgery outcomes.

PURPOSE: Maximum two-dimensional (2D) diameter has been used to define giant pituitary adenoma (GPA) surgery outcomes as has volume using an ellipsoid approximation of volumetrics. Cross sectional length can be measured in several different planes. We sought to compare the accuracy of different 2D cross sectional measurements with the 3D volumetric measurements for predicting GPA surgery outcomes. METHODS: Retrospective analysis was performed on a prospectively collected database. Tumors with >3 cm diameter were identified and classified based on maximal cross sectional measurements in three separate co-axial planes, i.e. transverse (TV), antero-posterior (AP) and cranio-caudal (CC). Volume was calculated using both MRI-guided volumetrics and an ellipsoid approximation (TV × AP × CC/2). Univariate and multivariate analysis was used to evaluate the relationship between cross sectional and volumetric data and extent of resection (EOR). RESULTS: In 62 subjects, median tumor volume using 3D volumetrics was 13.74 cm(3), which was overestimated by 16 % by the ellipsoid calculation (p = 0.0029), particularly for tumors >20 cm(3). Gross total resection (GTR) was 46.7 % and median EOR was 99.57 %. At 22-month follow-up, visual and anterior pituitary functions were stable (90 %) or improved (87 %). Pre-operative tumor volume >10 cm(3) (p = 0.02) and Knosp grade 3-4 (p = 0.04) were independent predictors of EOR. Knosp grade 3-4 (p < 0.0001), TV measurement >4 cm (p = 0.007) and maximum cross sectional length >4 cm (p = 0.04) were predictors of not achieving GTR. Only TV measurement (p = 0.02) predicted permanent diabetes insipidis. The smallest significant thresholds for predicting decreased GTR were TV measurement >25 mm, AP measurement >35 mm and volume >19 cm(3). CONCLUSION: We propose a new volumetric threshold of 20 cm(3) as most accurate for predicting GTR in the EEA era. CC measurement is the least useful predictor. Cavernous sinus invasion remains the best predictor of incomplete resection.

The role of ultrasonography in predicting vesicoureteral reflux.

OBJECTIVE: To evaluate the accuracy of renal and bladder ultrasonography (RBU) in predicting vesicoureteral reflux (VUR) in infants and children. MATERIALS AND METHODS: A total of 134 children who had VUR demonstrated on voiding cystourethrography (VCU) and also had RBU within 1 month of the VCU were included in the study, which took place between January 2005 and December 2012. VUR and hydronephrosis were graded with standard methods on VCU and RBU, respectively. Using VCU findings of reflux as the gold standard, diagnostic accuracy measures were performed for hydronephrosis and ureteral visualization on RBU, including sensitivity, specificity, positive predictive value, negative predictive value, and accuracy. RESULTS: Reflux grade was significantly associated with the degree of hydronephrosis (P = .0032). The sensitivity, negative predictive value, and accuracy of ultrasonography in predicting reflux was significantly higher for grade IV+ or grade V reflux compared with lower reflux grades. Also, the specificity of ultrasonography in predicting reflux was constant and at high level across all reflux grades, suggesting that ultrasonography is a good diagnostic screening tool. CONCLUSION: Normal RBU is rare with grade IV-V reflux, and moderate to severe hydronephrosis is rare with reflux grades

Utility of early postoperative high-resolution volumetric magnetic resonance imaging after transsphenoidal pituitary tumor surgery.

OBJECTIVE: Controversy exists over the utility of early postoperative magnetic resonance imaging (MRI) after transsphenoidal pituitary surgery for macroadenomas. We investigate whether valuable information can be derived from current greater resolution scans. METHODS: Volumetric MRI scans were obtained in the early (<10 days) and late (>30 days) postoperative periods in a series of patients undergoing transsphenoidal pituitary surgery. The volume of the residual tumor, resection cavity, and corresponding visual field tests were recorded at each time point. Statistical analyses of changes in tumor volume and cavity size were calculated using the late MRI as the gold standard. RESULTS: A total of 40 patients met the inclusion criteria. Preoperative tumor volume averaged 8.8 cm(3). Early postoperative assessment of average residual tumor volume (1.18 cm(3)) was quite accurate and did not differ statistically from late postoperative volume (1.23 cm(3), P = 0.64), indicating the utility of early scans to measure residual tumor. Early scans were 100% sensitive and 91% specific for predicting ≥ 98% resection (P < 0.001, Fisher exact test). The average percent decrease in cavity volume from preoperative MRI (tumor volume) to early postoperative imaging was 45% with decreases in all but 3 patients. There was no correlation between the size of the early cavity and the visual outcome. CONCLUSIONS: Early, high-resolution volumetric MRI is valuable in determining the presence or absence of residual tumor. Cavity volume almost always decreases after surgery, and a lack of decrease should alert the surgeon to possible persistent compression of the optic apparatus that may warrant reoperation.

Evaluation of the cardiac chambers on axial CT: comparison with echocardiography.

OBJECTIVE: To evaluate qualitative and simple quantitative measures of all 4 cardiac chamber sizes on computed tomography (CT) in comparison with transthoracic echocardiography (TTE). METHODS: We retrospectively identified 104 adults with electrocardiographically gated cardiac CT and TTE within 3 months. Axial early diastolic (75% R-R) CT images were reviewed for qualitative chamber enlargement, and each chamber was measured linearly. Transthoracic echocardiography was reviewed for linear, area, and volume measurements. Interrater agreement was calculated using Cohen κ and Pearson correlation. RESULTS: There were significant correlations between linear left atrium and left ventricle sizes by CT and TTE (r = 0.686 and r = 0.709, respectively). Correlations for right atrium and right ventricle measurements were lower (r = 0.447 and r = 0.492, respectively). Agreement between CT and TTE for qualitative chamber enlargement was poor (highest κ = 0.35). Computed tomography sensitivity was ≤ 62% for enlargement of all chambers. CONCLUSIONS: Linear CT measurements of left-sided chamber sizes correlate well with TTE. Right heart measurements and qualitative assessments agreed poorly with TTE.

Increased incidence of gastrointestinal bleeding following implantation of the HeartMate II LVAD.

BACKGROUND: The HeartMate II (HMII) Left Ventricular Assist System (Thoratec Corporation, Pleasanton, CA, USA), an axial continuous-flow left ventricular assist device (LVAD), has been approved for use in bridge-to-transplant patients and is under investigation for destination therapy. To avoid device-related thromboembolic complications, antiplatelet, and anticoagulation therapy are routinely administered. A worrisome frequency of gastrointestinal (GI) bleeding events has been observed. METHODS: A retrospective review of all 33 patients undergoing long-term LVAD implantation between June 1, 2006 and July 31, 2008 at our institution for any indication was conducted. Anticoagulation consisted of heparin (intravenous or subcutaneous) followed by transition to Coumadin therapy to a target INR of two to three. Antiplatelet therapy consisted of low-dose aspirin and dipyridamole. RESULTS: Twenty patients received the HMII and 13 patients received other devices. Eight (40%) HMII recipients suffered at least one episode of GI bleeding while no GI bleeding occurred in recipients of other devices (p = 0.012). Of 17 total bleeding episodes, no definitive source could be identified in 11 instances (65%). CONCLUSIONS: Although definitive source identification remains elusive, we believe that the majority of bleeding arises in the small bowel, possibly due to angiodysplasias, similar to the pathophysiology encountered in patients with aortic stenosis and GI bleeding. As we move toward wider use of the HMII and other axial continuous-flow devices in both bridge-to-transplant patients and for destination therapy, more studies will be necessary to understand the mechanisms of this obscure GI bleeding and develop treatment strategies to minimize its development.