Publications

Convolutional neural networks (CNN) have demonstrated good accuracy and speed in spatially registering high signal-to-noise ratio (SNR) structural magnetic resonance imaging (sMRI) images. However, some functional magnetic resonance imaging (fMRI) images, e.g., those acquired from arterial spin labeling (ASL) perfusion fMRI, are of intrinsically low SNR and therefore the quality of registering ASL images using CNN is not clear. In this work, we aimed to explore the feasibility of a CNN-based affine registration network (ARN) for registration of low-SNR three-dimensional ASL perfusion image time series and compare its performance with that from the state-of-the-art statistical parametric mapping (SPM) algorithm. The six affine parameters were learned from the ARN using both simulated motion and real acquisitions from ASL perfusion fMRI data and the registered images were generated by applying the transformation derived from the affine parameters. The speed and registration accuracy were compared between ARN and SPM. Several independent datasets, including meditation study (10 subjects × 2), bipolar disorder study (26 controls, 19 bipolar disorder subjects), and aging study (27 young subjects, 33 older subjects), were used to validate the generality of the trained ARN model. The ARN method achieves superior image affine registration accuracy (total translation/total rotation errors of ARN vs. SPM: 1.17 mm/1.23° vs. 6.09 mm/12.90° for simulated images and reduced MSE/L1/DSSIM/Total errors of 18.07% / 19.02% / 0.04% / 29.59% for real ASL test images) and 4.4 times (ARN vs. SPM: 0.50 s vs. 2.21 s) faster speed compared to SPM. The trained ARN can be generalized to align ASL perfusion image time series acquired with different scanners, and from different image resolutions, and from healthy or diseased populations. The results demonstrated that our ARN markedly outperforms the iteration-based SPM both for simulated motion and real acquisitions in terms of registration accuracy, speed, and generalization.

OBJECTIVES: To determine the factors that affect successful ultrasound-guided biopsy of liver lesions and build a model predicting feasibility of US-guided liver biopsy.

METHODS: This is IRB-approved HIPAA-compliant retrospective review of consecutive ultrasound-guided targeted liver biopsies performed or attempted between 1/2018 and 9/2020 at a single tertiary academic institution with a total of 501 patients included. Mann-Whitney and chi-square tests were used to compare continuous and categorical variables, respectively. Logistic regression model was built to predict feasibility of successful ultrasound-guided biopsy.

RESULTS: Liver lesion biopsy was successfully performed with US guidance in 429/501 (86%) patients. Lesions not amenable for US biopsy were smaller (median size 1.6 cm vs 3.3 cm, p < 0.0001) and deeper within the liver (median depth 9.0 cm vs 5.8 cm, p < 0.0001). The technical success rate was lowest for lesions in segment II (40/53, 75%), while lesions in segment IVb (87/91, 96%) had highest success rate (p < 0.003). US targeting in patients with 1 or 2 lesions was less feasible than in patients with 3 or more lesions, 126/180 (70%) vs. 303/321 (94%), (p < 0.0001). Model including lesion size, depth, location, and number of lesions predicts feasibility of US-guided biopsy with Area under the ROC curve (AUC) = 0.92.

CONCLUSIONS: Linear logistic regression model that includes lesion size, depth and location, and number of lesions is highly successful in predicting feasibility of ultrasound-guided biopsy for liver lesions. Smaller lesions, deeper lesions, and lesions in segment II and VIII in patients with less than 3 lesions were less feasible for ultrasound-guided biopsy of liver lesions.

This study aimed to evaluate the geographic patient profile of a country's first interventional radiology (IR) service in sub-Saharan Africa. From October 2018 to August 2022, travel time (1,339 patients) and home region (1,184 patients) were recorded from 1,434 patients who underwent IR procedures at Tanzania's largest referral center. Distances traveled by road were calculated from the administrative capital of each region using a web mapping platform (google.com/maps). The effect of various factors on distance and time traveled were assessed. Patients from all 31 regions in Tanzania underwent IR procedures. The mean and maximum calculated distance traveled by patients were 241.6 km and 1,387 km, respectively (Sk2 = 1.66); 25.0% of patients traveled for over 6 hours for their procedure. Patients traveled furthest for genitourinary procedures (mean = 293.4 km) and least for angioplasty and stent placement (mean = 123.9 km) (P < .001). To increase population access and reduce travel times, geographic data should be used to decentralize services.

Registry data are being increasingly used to establish treatment guidelines, set benchmarks, allocate resources, and make payment decisions. Although many registries rely on manual data entry, the Society of Interventional Radiology (SIR) is using automated data extraction for its VIRTEX registry. This process relies on participants using consistent terminology with highly structured data in physician-developed standardized reports (SR). To better understand barriers to adoption, a survey was sent to 3,178 SIR members. Responses were obtained from 451 interventional radiology practitioners (14.2%) from 92 unique academic and 151 unique private practices. Of these, 75% used structured reports and 32% used the SIR SR. The most common barriers to the use of these reports include SR length (35% of respondents), lack of awareness about the SR (31%), and lack of agreement on adoption within practices (27%). The results demonstrated insights regarding barriers in the use and/or adoption of SR and potential solutions.

PURPOSE: Chiari malformation type I (CMI) patients have been independently shown to have both increased resistance to cerebrospinal fluid (CSF) flow in the cervical spinal canal and greater cardiac-induced neural tissue motion compared to healthy controls. The goal of this paper is to determine if a relationship exists between CSF flow resistance and brain tissue motion in CMI subjects.

METHODS: Computational fluid dynamics (CFD) techniques were employed to compute integrated longitudinal impedance (ILI) as a measure of unsteady resistance to CSF flow in the cervical spinal canal in thirty-two CMI subjects and eighteen healthy controls. Neural tissue motion during the cardiac cycle was assessed using displacement encoding with stimulated echoes (DENSE) magnetic resonance imaging (MRI) technique.

RESULTS: The results demonstrate a positive correlation between resistance to CSF flow and the maximum displacement of the cerebellum for CMI subjects (r = 0.75, p = 6.77 × 10-10) but not for healthy controls. No correlation was found between CSF flow resistance and maximum displacement in the brainstem for CMI or healthy subjects. The magnitude of resistance to CSF flow and maximum cardiac-induced brain tissue motion were not statistically different for CMI subjects with and without the presence of five CMI symptoms: imbalance, vertigo, swallowing difficulties, nausea or vomiting, and hoarseness.

CONCLUSION: This study establishes a relationship between CSF flow resistance in the cervical spinal canal and cardiac-induced brain tissue motion in the cerebellum for CMI subjects. Further research is necessary to understand the importance of resistance and brain tissue motion in the symptomatology of CMI.

Convolutional neural networks (CNN) have demonstrated good accuracy and speed in spatially registering high signal-to-noise ratio (SNR) structural magnetic resonance imaging (sMRI) images. However, some functional magnetic resonance imaging (fMRI) images, e.g., those acquired from arterial spin labeling (ASL) perfusion fMRI, are of intrinsically low SNR and therefore the quality of registering ASL images using CNN is not clear. In this work, we aimed to explore the feasibility of a CNN-based affine registration network (ARN) for registration of low-SNR three-dimensional ASL perfusion image time series and compare its performance with that from the state-of-the-art statistical parametric mapping (SPM) algorithm. The six affine parameters were learned from the ARN using both simulated motion and real acquisitions from ASL perfusion fMRI data and the registered images were generated by applying the transformation derived from the affine parameters. The speed and registration accuracy were compared between ARN and SPM. Several independent datasets, including meditation study (10 subjects × 2), bipolar disorder study (26 controls, 19 bipolar disorder subjects), and aging study (27 young subjects, 33 older subjects), were used to validate the generality of the trained ARN model. The ARN method achieves superior image affine registration accuracy (total translation/total rotation errors of ARN vs. SPM: 1.17 mm/1.23° vs. 6.09 mm/12.90° for simulated images and reduced MSE/L1/DSSIM/Total errors of 18.07% / 19.02% / 0.04% / 29.59% for real ASL test images) and 4.4 times (ARN vs. SPM: 0.50 s vs. 2.21 s) faster speed compared to SPM. The trained ARN can be generalized to align ASL perfusion image time series acquired with different scanners, and from different image resolutions, and from healthy or diseased populations. The results demonstrated that our ARN markedly outperforms the iteration-based SPM both for simulated motion and real acquisitions in terms of registration accuracy, speed, and generalization.

BACKGROUND: Multidisciplinary orthopaedic oncology conferences are important in developing the treatment plan for patients with suspected orthopaedic bone and soft tissue tumors, involving physicians from several services. Past studies have shown the clinical value of these conferences; however, the impact of radiology input on the management plan and time cost for radiology to staff these conferences has not been fully studied.

QUESTIONS/PURPOSES: (1) Does radiology input at multidisciplinary conference help guide clinical management and improve clinician confidence? (2) What is the time cost of radiology input for a multidisciplinary conference?

METHODS: This prospective study was conducted from October 2020 to March 2022 at a tertiary academic center with a sarcoma center. A single data questionnaire for each patient was sent to one of three treating orthopaedic oncologists with 41, 19, and 5 years of experience after radiology discussion at a weekly multidisciplinary conference. A data questionnaire was completed by the treating orthopaedic oncologist for 48% (322 of 672) of patients, which refers to the proportion of those three oncologists' patients for which survey data were captured. A musculoskeletal radiology fellow and musculoskeletal fellowship-trained radiology attending physician provided radiology input at each multidisciplinary conference. The clinical plan (leave alone, follow-up imaging, follow-up clinically, recommend different imaging test, core needle biopsy, surgical excision or biopsy or fixation, or other) and change in clinical confidence before and after radiology input were documented. A second weekly data questionnaire was sent to the radiology fellow to estimate the time cost of radiology input for the multidisciplinary conference.

RESULTS: In 29% (93 of 322) of patients, there was a change in the clinical plan after radiology input. Biopsy was canceled in 30% (24 of 80) of patients for whom biopsy was initially planned, and surgical excision was canceled in 24% (17 of 72) of patients in whom surgical excision was initially planned. In 21% (68 of 322) of patients, there were unreported imaging findings that affected clinical management; 13% (43 of 322) of patients had a missed finding, and 8% (25 of 322) of patients had imaging findings that were interpreted incorrectly. For confidence in the final treatment plan, 78% (251 of 322) of patients had an increase in clinical confidence by their treating orthopaedic oncologist after the multidisciplinary conference. Radiology fellows and attendings spent a mean of 4.2 and 1.5 hours, respectively, reviewing and presenting at a multidisciplinary conference each week. The annual combined prorated time cost for the radiology attending and fellow was estimated at USD 24,310 based on national median salary data for attendings and internal salary data for fellows.

CONCLUSION: In a study taken at one tertiary-care oncology program, input from radiology attendings and fellows in the setting of a multidisciplinary conference helped to guide the final treatment plan, reduce procedures, and improve clinician confidence in the final treatment plan, at an annual time cost of USD 24,310.

CLINICAL RELEVANCE: Multidisciplinary orthopaedic oncology conferences can lead to changes in management plans, and the time cost to the radiologists should be budgeted for by the radiology department or parent institution.

PURPOSE: To assess the accuracy, completeness, and readability of patient educational material produced by a machine learning model and compare the output to that provided by a societal website.

MATERIALS AND METHODS: Content from the Society of Interventional Radiology Patient Center website was retrieved, categorized, and organized into discrete questions. These questions were entered into the ChatGPT platform, and the output was analyzed for word and sentence counts, readability using multiple validated scales, factual correctness, and suitability for patient education using the Patient Education Materials Assessment Tool for Printable Materials (PEMAT-P) instrument.

RESULTS: A total of 21,154 words were analyzed, including 7,917 words from the website and 13,377 words representing the total output of the ChatGPT platform across 22 text passages. Compared to the societal website, output from the ChatGPT platform was longer and more difficult to read on 4 of 5 readability scales. The ChatGPT output was incorrect for 12 (11.5%) of 104 questions. When reviewed using the PEMAT-P tool, the ChatGPT content scored lower than the website material. Content from both the website and ChatGPT were significantly above the recommended fifth or sixth grade level for patient education, with a mean Flesch-Kincaid grade level of 11.1 (±1.3) for the website and 11.9 (±1.6) for the ChatGPT content.

CONCLUSIONS: The ChatGPT platform may produce incomplete or inaccurate patient educational content, and providers should be familiar with the limitations of the system in its current form. Opportunities may exist to fine-tune existing large language models, which could be optimized for the delivery of patient educational content.

Lung cancer continues to be the most common cause of cancer-related death worldwide. In the past decade, with the implementation of lung cancer screening programs and advances in surgical and nonsurgical therapies, the survival of patients with lung cancer has increased, as has the number of imaging studies that these patients undergo. However, most patients with lung cancer do not undergo surgical re-section, because they have comorbid disease or lung cancer in an advanced stage at diagnosis. Nonsurgical therapies have continued to evolve with a growing range of systemic and targeted therapies, and there has been an associated evolution in the imaging findings encountered at follow-up examinations after such therapies (e.g., with respect to posttreatment changes, treatment complications, and recurrent tumor). This AJR Expert Panel Narrative Review describes the current status of nonsurgical therapies for lung cancer and their expected and unexpected imaging manifestations. The goal is to provide guidance to radiologists regarding imaging assessment after such therapies, focusing mainly on non-small cell lung cancer. Covered therapies include systemic therapy (conventional chemotherapy, targeted therapy, and immunotherapy), radiotherapy, and thermal ablation.