Using deep learning-derived image features in radiologic time series to make personalised predictions: proof of concept in colonic transit data.

OBJECTIVES: Siamese neural networks (SNN) were used to classify the presence of radiopaque beads as part of a colonic transit time study (CTS). The SNN output was then used as a feature in a time series model to predict progression through a CTS. METHODS: This retrospective study included all patients undergoing a CTS in a single institution from 2010 to 2020. Data were partitioned in an 80/20 Train/Test split. Deep learning models based on a SNN architecture were trained and tested to classify images according to the presence, absence, and number of radiopaque beads and to output the Euclidean distance between the feature representations of the input images. Time series models were used to predict the total duration of the study. RESULTS: In total, 568 images of 229 patients (143, 62% female, mean age 57) patients were included. For the classification of the presence of beads, the best performing model (Siamese DenseNET trained with a contrastive loss with unfrozen weights) achieved an accuracy, precision, and recall of 0.988, 0.986, and 1. A Gaussian process regressor (GPR) trained on the outputs of the SNN outperformed both GPR using only the number of beads and basic statistical exponential curve fitting with MAE of 0.9 days compared to 2.3 and 6.3 days (p < 0.05) respectively. CONCLUSIONS: SNNs perform well at the identification of radiopaque beads in CTS. For time series prediction our methods were superior at identifying progression through the time series compared to statistical models, enabling more accurate personalised predictions. CLINICAL RELEVANCE STATEMENT: Our radiologic time series model has potential clinical application in use cases where change assessment is critical (e.g. nodule surveillance, cancer treatment response, and screening programmes) by quantifying change and using it to make more personalised predictions. KEY POINTS: • Time series methods have improved but application to radiology lags behind computer vision. Colonic transit studies are a simple radiologic time series measuring function through serial radiographs. • We successfully employed a Siamese neural network (SNN) to compare between radiographs at different points in time and then used the output of SNN as a feature in a Gaussian process regression model to predict progression through the time series. • This novel use of features derived from a neural network on medical imaging data to predict progression has potential clinical application in more complex use cases where change assessment is critical such as in oncologic imaging, monitoring for treatment response, and screening programmes.

Cybersecurity considerations for radiology departments involved with artificial intelligence.

Radiology artificial intelligence (AI) projects involve the integration of integrating numerous medical devices, wireless technologies, data warehouses, and social networks. While cybersecurity threats are not new to healthcare, their prevalence has increased with the rise of AI research for applications in radiology, making them one of the major healthcare risks of 2021. Radiologists have extensive experience with the interpretation of medical imaging data but radiologists may not have the required level of awareness or training related to AI-specific cybersecurity concerns. Healthcare providers and device manufacturers can learn from other industry sector industries that have already taken steps to improve their cybersecurity systems. This review aims to introduce cybersecurity concepts as it relates to medical imaging and to provide background information on general and healthcare-specific cybersecurity challenges. We discuss approaches to enhancing the level and effectiveness of security through detection and prevention techniques, as well as ways that technology can improve security while mitigating risks. We first review general cybersecurity concepts and regulatory issues before examining these topics in the context of radiology AI, with a specific focus on data, training, data, training, implementation, and auditability. Finally, we suggest potential risk mitigation strategies. By reading this review, healthcare providers, researchers, and device developers can gain a better understanding of the potential risks associated with radiology AI projects, as well as strategies to improve cybersecurity and reduce potential associated risks. CLINICAL RELEVANCE STATEMENT: This review can aid radiologists' and related professionals' understanding of the potential cybersecurity risks associated with radiology AI projects, as well as strategies to improve security. KEY POINTS: • Embarking on a radiology artificial intelligence (AI) project is complex and not without risk especially as cybersecurity threats have certainly become more abundant in the healthcare industry. • Fortunately healthcare providers and device manufacturers have the advantage of being able to take inspiration from other industry sectors who are leading the way in the field. • Herein we provide an introduction to cybersecurity as it pertains to radiology, a background to both general and healthcare-specific cybersecurity challenges; we outline general approaches to improving security through both detection and preventative techniques, and instances where technology can increase security while mitigating risks.

Radiology artificial intelligence: a systematic review and evaluation of methods (RAISE).

OBJECTIVE: There has been a large amount of research in the field of artificial intelligence (AI) as applied to clinical radiology. However, these studies vary in design and quality and systematic reviews of the entire field are lacking.This systematic review aimed to identify all papers that used deep learning in radiology to survey the literature and to evaluate their methods. We aimed to identify the key questions being addressed in the literature and to identify the most effective methods employed. METHODS: We followed the PRISMA guidelines and performed a systematic review of studies of AI in radiology published from 2015 to 2019. Our published protocol was prospectively registered. RESULTS: Our search yielded 11,083 results. Seven hundred sixty-seven full texts were reviewed, and 535 articles were included. Ninety-eight percent were retrospective cohort studies. The median number of patients included was 460. Most studies involved MRI (37%). Neuroradiology was the most common subspecialty. Eighty-eight percent used supervised learning. The majority of studies undertook a segmentation task (39%). Performance comparison was with a state-of-the-art model in 37%. The most used established architecture was UNet (14%). The median performance for the most utilised evaluation metrics was Dice of 0.89 (range .49-.99), AUC of 0.903 (range 1.00-0.61) and Accuracy of 89.4 (range 70.2-100). Of the 77 studies that externally validated their results and allowed for direct comparison, performance on average decreased by 6% at external validation (range increase of 4% to decrease 44%). CONCLUSION: This systematic review has surveyed the major advances in AI as applied to clinical radiology. KEY POINTS: • While there are many papers reporting expert-level results by using deep learning in radiology, most apply only a narrow range of techniques to a narrow selection of use cases. • The literature is dominated by retrospective cohort studies with limited external validation with high potential for bias. • The recent advent of AI extensions to systematic reporting guidelines and prospective trial registration along with a focus on external validation and explanations show potential for translation of the hype surrounding AI from code to clinic.

Risk Stratification in Multinodular Goiter: A Retrospective Review of Sonographic Features, Histopathological Results, and Cancer Risk.

PURPOSE: In the management of thyroid nodules, although the potential for malignancy exists, there is also the potential for overtreatment of subclinical disease. Although the TI-RADS (Thyroid Imaging-Reporting and Data System) system outlines a risk stratification score based on suspicious ultrasound findings, it has not been universally accepted. Many TI-RADS 2 or 3 patients proceed to fine needle aspiration biopsy (FNAB), potentially unnecessarily. The aim of the study was to identify whether lesions within a multinodular goiter (MNG) without suspicious features can be followed with ultrasound rather than biopsied as is recommended for single nodules. METHODS: Pathology records were retrospectively analysed for proven MNGs over a 5-year period. A total of 293 cases were identified. FNAB, prebiopsy ultrasound images, and reports were identified for each case. Images were reviewed and assessed for sonographically suspicious criteria guided by TI-RADS. Logistic regression was applied to determine if any sonographic features were associated with neoplasia. Odds ratios with 95% confidence intervals were calculated. RESULTS: Of 293 samples, 14 (4.7%) were neoplastic. Having no suspicious features conferred an average risk of 0.0339 (95% confidence interval: 0.02831-0.04087) of neoplasia. Risk of neoplasm significantly increased by having 1 and >1 suspicious feature (P < .001). Regarding cytological results, of 237 patients with Thy-2 cytology, 159 were followed up and 8 had a neoplasm. CONCLUSION: Ultrasound can be used to estimate risk of neoplasia in MNG. In the absence of suspicious radiological findings, follow-up with ultrasound rather than FNAB may be appropriate in patients who have a low clinical suspicion for neoplasia.

The Development of Expertise in Radiology: In Chest Radiograph Interpretation, "Expert" Search Pattern May Predate "Expert" Levels of Diagnostic Accuracy for Pneumothorax Identification.

Purpose To investigate the development of chest radiograph interpretation skill through medical training by measuring both diagnostic accuracy and eye movements during visual search. Materials and Methods An institutional exemption from full ethical review was granted for the study. Five consultant radiologists were deemed the reference expert group, and four radiology registrars, five senior house officers (SHOs), and six interns formed four clinician groups. Participants were shown 30 chest radiographs, 14 of which had a pneumothorax, and were asked to give their level of confidence as to whether a pneumothorax was present. Receiver operating characteristic (ROC) curve analysis was carried out on diagnostic decisions. Eye movements were recorded with a Tobii TX300 (Tobii Technology, Stockholm, Sweden) eye tracker. Four eye-tracking metrics were analyzed. Variables were compared to identify any differences between groups. All data were compared by using the Friedman nonparametric method. Results The average area under the ROC curve for the groups increased with experience (0.947 for consultants, 0.792 for registrars, 0.693 for SHOs, and 0.659 for interns; P = .009). A significant difference in diagnostic accuracy was found between consultants and registrars (P = .046). All four eye-tracking metrics decreased with experience, and there were significant differences between registrars and SHOs. Total reading time decreased with experience; it was significantly lower for registrars compared with SHOs (P = .046) and for SHOs compared with interns (P = .025). Conclusion Chest radiograph interpretation skill increased with experience, both in terms of diagnostic accuracy and visual search. The observed level of experience at which there was a significant difference was higher for diagnostic accuracy than for eye-tracking metrics. (©) RSNA, 2016 Online supplemental material is available for this article.

A Radiomic "Warning Sign" of Progression on Brain MRI in Individuals with MS.

BACKGROUND AND PURPOSE: MS is a chronic progressive, idiopathic, demyelinating disorder whose diagnosis is contingent on the interpretation of MR imaging. New MR imaging lesions are an early biomarker of disease progression. We aimed to evaluate a machine learning model based on radiomics features in predicting progression on MR imaging of the brain in individuals with MS. MATERIALS AND METHODS: This retrospective cohort study with external validation on open-access data obtained full ethics approval. Longitudinal MR imaging data for patients with MS were collected and processed for machine learning. Radiomics features were extracted at the future location of a new lesion in the patients' prior MR imaging ("prelesion"). Additionally, "control" samples were obtained from the normal-appearing white matter for each participant. Machine learning models for binary classification were trained and tested and then evaluated the external data of the model. RESULTS: The total number of participants was 167. Of the 147 in the training/test set, 102 were women and 45 were men. The average age was 42 (range, 21-74 years). The best-performing radiomics-based model was XGBoost, with accuracy, precision, recall, and F1-score of 0.91, 0.91, 0.91, and 0.91 on the test set, and 0.74, 0.74, 0.74, and 0.70 on the external validation set. The 5 most important radiomics features to the XGBoost model were associated with the overall heterogeneity and low gray-level emphasis of the segmented regions. Probability maps were produced to illustrate potential future clinical applications. CONCLUSIONS: Our machine learning model based on radiomics features successfully differentiated prelesions from normal-appearing white matter. This outcome suggests that radiomics features from normal-appearing white matter could serve as an imaging biomarker for progression of MS on MR imaging.

iSPAN: Explainable prediction of outcomes post thrombectomy with Machine Learning.

PURPOSE: This study aimed to develop and evaluate a machine learning model and a novel clinical score for predicting outcomes in stroke patients undergoing endovascular thrombectomy. MATERIALS AND METHODS: This retrospective study included all patients aged over 18 years with an anterior circulation stroke treated at a thrombectomy centre from 2010 to 2020 with external validation. The primary outcome was day 90 mRS ≥3. Existing clinical scores (SPAN and PRE) and Machine Learning (ML) models were compared. A novel clinical score (iSPAN) was derived by adding an optimised weighting of the most important ML features to the SPAN. RESULTS: 812 patients were initially included (397 female, average age 73), 63 for external validation. The best performing clinical score and ML model were SPAN and XGB (sensitivity, specificity and accuracy 0.290, 0.967, 0.628 and 0.693, 0.783, 0.738 respectively). A significant difference was found overall and our XGB model was more accurate than SPAN (p < 0.0018). The most important features were Age, mTICI and total number of passes. The addition of 11 points for mTICI of ≤2B and 3 points for ≥3 passes to the SPAN achieved the best accuracy and was used to create the iSPAN. iSPAN was not significantly less accurate than our XGB model (p > 0.5). In the external validation set, iSPAN and SPAN achieved sensitivity, specificity, and accuracy of (0.735, 0.862, 0.79) and (0.471, 0.897, 0.67) respectively. CONCLUSION: iSPAN incorporates machine-derived features to achieve better predictions compared to existing clinical scores. It is not inferior to our XGB model and is externally generalisable.

METhodological RadiomICs Score (METRICS): a quality scoring tool for radiomics research endorsed by EuSoMII.

PURPOSE: To propose a new quality scoring tool, METhodological RadiomICs Score (METRICS), to assess and improve research quality of radiomics studies. METHODS: We conducted an online modified Delphi study with a group of international experts. It was performed in three consecutive stages: Stage#1, item preparation; Stage#2, panel discussion among EuSoMII Auditing Group members to identify the items to be voted; and Stage#3, four rounds of the modified Delphi exercise by panelists to determine the items eligible for the METRICS and their weights. The consensus threshold was 75%. Based on the median ranks derived from expert panel opinion and their rank-sum based conversion to importance scores, the category and item weights were calculated. RESULT: In total, 59 panelists from 19 countries participated in selection and ranking of the items and categories. Final METRICS tool included 30 items within 9 categories. According to their weights, the categories were in descending order of importance: study design, imaging data, image processing and feature extraction, metrics and comparison, testing, feature processing, preparation for modeling, segmentation, and open science. A web application and a repository were developed to streamline the calculation of the METRICS score and to collect feedback from the radiomics community. CONCLUSION: In this work, we developed a scoring tool for assessing the methodological quality of the radiomics research, with a large international panel and a modified Delphi protocol. With its conditional format to cover methodological variations, it provides a well-constructed framework for the key methodological concepts to assess the quality of radiomic research papers. CRITICAL RELEVANCE STATEMENT: A quality assessment tool, METhodological RadiomICs Score (METRICS), is made available by a large group of international domain experts, with transparent methodology, aiming at evaluating and improving research quality in radiomics and machine learning. KEY POINTS: • A methodological scoring tool, METRICS, was developed for assessing the quality of radiomics research, with a large international expert panel and a modified Delphi protocol. • The proposed scoring tool presents expert opinion-based importance weights of categories and items with a transparent methodology for the first time. • METRICS accounts for varying use cases, from handcrafted radiomics to entirely deep learning-based pipelines. • A web application has been developed to help with the calculation of the METRICS score ( https://metricsscore.github.io/metrics/METRICS.html ) and a repository created to collect feedback from the radiomics community ( https://github.com/metricsscore/metrics ).

How to apply evidence-based practice to the use of artificial intelligence in radiology (EBRAI) using the data algorithm training output (DATO) method.

OBJECTIVE: As the number of radiology artificial intelligence (AI) papers increases, there are new challenges for reviewing the AI literature as well as differences to be aware of, for those familiar with the clinical radiology literature. We aim to introduce a tool to aid in this process. METHODS: In evidence-based practise (EBP), you must Ask, Search, Appraise, Apply and Evaluate to come to an evidence-based decision. The bottom-up evidence-based radiology (EBR) method allows for a systematic way of choosing the correct radiological investigation or treatment. Just as the population intervention comparison outcome (PICO) method is an established means of asking an answerable question; herein, we introduce the data algorithm training output (DATO) method to complement PICO by considering Data, Algorithm, Training and Output in the use of AI to answer the question. RESULTS: We illustrate the DATO method with a worked example concerning bone age assessment from skeletal radiographs. After a systematic search, 17 bone age estimation papers (5 of which externally validated their results) were appraised. The paper with the best DATO metrics found that an ensemble model combining uncorrelated, high performing simple models should achieve error rates comparable to human performance. CONCLUSION: Considering DATO in the application of EBR to AI is a simple systematic approach to this potentially daunting subject. ADVANCES IN KNOWLEDGE: The growth of AI in radiology means that radiologists and related professionals now need to be able to review not only clinical radiological literature but also research using AI methods. Considering Data, Algorithm, Training and Output in the application of EBR to AI is a simple systematic approach to this potentially daunting subject.

Metformin Abrogates Age-Associated Ovarian Fibrosis.

PURPOSE: The ovarian cancer risk factors of age and ovulation are curious because ovarian cancer incidence increases in postmenopausal women, long after ovulations have ceased. To determine how age and ovulation underlie ovarian cancer risk, we assessed the effects of these risk factors on the ovarian microenvironment. EXPERIMENTAL DESIGN: Aged C57/lcrfa mice (0-33 months old) were generated to assess the aged ovarian microenvironment. To expand our findings into human aging, we assembled a cohort of normal human ovaries (n = 18, 21-71 years old). To validate our findings, an independent cohort of normal human ovaries was assembled (n = 9, 41-82 years old). RESULTS: We first validated the presence of age-associated murine ovarian fibrosis. Using interdisciplinary methodologies, we provide novel evidence that ovarian fibrosis also develops in human postmenopausal ovaries across two independent cohorts (n = 27). Fibrotic ovaries have an increased CD206+:CD68+ cell ratio, CD8+ T-cell infiltration, and profibrotic DPP4+αSMA+ fibroblasts. Metformin use was associated with attenuated CD8+ T-cell infiltration and reduced CD206+:CD68+ cell ratio. CONCLUSIONS: These data support a novel hypothesis that unifies the primary nonhereditary ovarian cancer risk factors through the development of ovarian fibrosis and the formation of a premetastatic niche, and suggests a potential use for metformin in ovarian cancer prophylaxis.See related commentary by Madariaga et al., p. 523.

Breast Arterial Calcification on screening mammography can predict significant Coronary Artery Disease in women.

INTRODUCTION: Breast Arterial Calcification (BAC) on digital mammography has been associated with an increased risk of Coronary Artery Disease (CAD). We aimed to investigate the association of BAC with findings on Coronary Computed Tomography Angiography (CCTA) within a cohort of women from the national breast screening program. METHODS: Symptomatic women (chest pain) aged between 50 and 65 who underwent a CCTA and who also had a screening mammography between 2014 and 2015 were recorded. BAC and CAD-RADS™: Coronary Artery Disease-Reporting and Data System were scored by separate blinded specialist radiologists. Cardiac risk factors were recorded. Patients' cardiac follow up (with Exercise Stress Test, Percutaneous Coronary Intervention or echocardiography) and cardio-protective medications were also documented. RESULTS: 219 eligible women underwent a CCTA. Of these, 104 patients also underwent digital mammography. Using standard linear regression BAC was identified as a significant predictor of CAD-RADs ≥3 disease. Using binomial logistic regression, BAC remained associated with CAD-RADs ≥3 (p=0.023). A significantly higher proportion of patients with BAC >1 were on cardio-protective medications (p=0.041) and had medications initiated or changed, or had further cardiac investigation (p=0.037 and p=0.019, respectively) than those with no BAC, after a mean follow-up of 20.6 (range 15-27) months. CONCLUSION: BAC diagnosed on 2 yearly screening mammography predicts CAD-RADs ≥3 disease in symptomatic patients.

The Use of Twitter by Radiology Journals: An Analysis of Twitter Activity and Impact Factor.

PURPOSE: Medical journals use social media as a means to disseminate new research and interact with readers. The microblogging site Twitter is one such platform. The aim of this study was to analyze the recent use of Twitter by the leading radiology journals. METHODS: The top 50 journals by Impact Factor were included. Twitter profiles associated with these journals, or their corresponding societies, were identified. Whether each journal used other social media platforms was also recorded. Each Twitter profile was analyzed over a one-year period, with data collected via Twitonomy software. Klout scores of social media influence were calculated. Results were analyzed in SPSS using Student's t test, Fisher contingency tables, and Pearson correlations to identify any association between social media interaction and Impact Factors of journals. RESULTS: Fourteen journals (28%) had dedicated Twitter profiles. Of the 36 journals without dedicated Twitter profiles, 25 (50%) were associated with societies that had profiles, leaving 11 (22%) journals without a presence on Twitter. The mean Impact Factor of all journals was 3.1 ± 1.41 (range, 1.7-6.9). Journals with Twitter profiles had higher Impact Factors than those without (mean, 3.37 vs 2.14; P < .001). There was no statistically significant difference between the Impact Factors of the journals with dedicated Twitter profiles and those associated with affiliated societies (P = .47). Since joining Twitter, 7 of the 11 journals (64%) experienced increases in Impact Factor. A greater number of Twitter followers was correlated with higher journal Impact Factor (R2 = 0.581, P = .029). CONCLUSIONS: The investigators assessed the prevalence and activity of the leading radiology journals on Twitter. Radiology journals with Twitter profiles have higher Impact Factors than those without profiles, and the number of followers of a journal's Twitter profile is positively associated with Impact Factor.