Random expert sampling for deep learning segmentation of acute ischemic stroke on non-contrast CT.

BACKGROUND: Outlining acutely infarcted tissue on non-contrast CT is a challenging task for which human inter-reader agreement is limited. We explored two different methods for training a supervised deep learning algorithm: one that used a segmentation defined by majority vote among experts and another that trained randomly on separate individual expert segmentations. METHODS: The data set consisted of 260 non-contrast CT studies in 233 patients with acute ischemic stroke recruited from the multicenter DEFUSE 3 (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke 3) trial. Additional external validation was performed using 33 patients with matched stroke onset times from the University Hospital Lausanne. A benchmark U-Net was trained on the reference annotations of three experienced neuroradiologists to segment ischemic brain tissue using majority vote and random expert sampling training schemes. The median of volume, overlap, and distance segmentation metrics were determined for agreement in lesion segmentations between (1) three experts, (2) the majority model and each expert, and (3) the random model and each expert. The two sided Wilcoxon signed rank test was used to compare performances (1) to 2) and (1) to (3). We further compared volumes with the 24 hour follow-up diffusion weighted imaging (DWI, final infarct core) and correlations with clinical outcome (modified Rankin Scale (mRS) at 90 days) with the Spearman method. RESULTS: The random model outperformed the inter-expert agreement ((1) to (2)) and the majority model ((1) to (3)) (dice 0.51±0.04 vs 0.36±0.05 (P<0.0001) vs 0.45±0.05 (P<0.0001)). The random model predicted volume correlated with clinical outcome (0.19, P<0.05), whereas the median expert volume and majority model volume did not. There was no significant difference when comparing the volume correlations between random model, median expert volume, and majority model to 24 hour follow-up DWI volume (P>0.05, n=51). CONCLUSION: The random model for ischemic injury delineation on non-contrast CT surpassed the inter-expert agreement ((1) to (2)) and the performance of the majority model ((1) to (3)). We showed that the random model volumetric measures of the model were consistent with 24 hour follow-up DWI.

Pretrained Neural Networks Accurately Identify Cancer Recurrence in Medical Record.

Cancer recurrence is the diagnosis of a second clinical episode of cancer after the first was considered cured. Identifying patients who had experienced cancer recurrence is an important task as it can be used to compare treatment effectiveness, measure recurrence-free survival, and plan and prioritize cancer control resources. We developed BERT-based natural language processing (NLP) contextual models for identifying cancer recurrence incidence and the recurrence time based on the records in progress notes. Using two datasets containing breast and colorectal cancer patients, we demonstrated the advantage of the contextual models over the traditional NLP models by overcoming the laborious and often unscalable tasks of composing keywords in a specific disease domain.

Artificial Intelligence and Deep Learning in Neuroradiology: Exploring the New Frontier.

There have been many recently published studies exploring machine learning (ML) and deep learning applications within neuroradiology. The improvement in performance of these techniques has resulted in an ever-increasing number of commercially available tools for the neuroradiologist. In this narrative review, recent publications exploring ML in neuroradiology are assessed with a focus on several key clinical domains. In particular, major advances are reviewed in the context of: (1) intracranial hemorrhage detection, (2) stroke imaging, (3) intracranial aneurysm screening, (4) multiple sclerosis imaging, (5) neuro-oncology, (6) head and tumor imaging, and (7) spine imaging.

netDx: Software for building interpretable patient classifiers by multi-'omic data integration using patient similarity networks.

Patient classification based on clinical and genomic data will further the goal of precision medicine. Interpretability is of particular relevance for models based on genomic data, where sample sizes are relatively small (in the hundreds), increasing overfitting risk netDx is a machine learning method to integrate multi-modal patient data and build a patient classifier. Patient data are converted into networks of patient similarity, which is intuitive to clinicians who also use patient similarity for medical diagnosis. Features passing selection are integrated, and new patients are assigned to the class with the greatest profile similarity. netDx has excellent performance, outperforming most machine-learning methods in binary cancer survival prediction. It handles missing data - a common problem in real-world data - without requiring imputation. netDx also has excellent interpretability, with native support to group genes into pathways for mechanistic insight into predictive features. The netDx Bioconductor package provides multiple workflows for users to build custom patient classifiers. It provides turnkey functions for one-step predictor generation from multi-modal data, including feature selection over multiple train/test data splits. Workflows offer versatility with custom feature design, choice of similarity metric; speed is improved by parallel execution. Built-in functions and examples allow users to compute model performance metrics such as AUROC, AUPR, and accuracy. netDx uses RCy3 to visualize top-scoring pathways and the final integrated patient network in Cytoscape. Advanced users can build more complex predictor designs with functional building blocks used in the default design. Finally, the netDx Bioconductor package provides a novel workflow for pathway-based patient classification from sparse genetic data.

netDx: interpretable patient classification using integrated patient similarity networks.

Patient classification has widespread biomedical and clinical applications, including diagnosis, prognosis, and treatment response prediction. A clinically useful prediction algorithm should be accurate, generalizable, be able to integrate diverse data types, and handle sparse data. A clinical predictor based on genomic data needs to be interpretable to drive hypothesis-driven research into new treatments. We describe netDx, a novel supervised patient classification framework based on patient similarity networks, which meets these criteria. In a cancer survival benchmark dataset integrating up to six data types in four cancer types, netDx significantly outperforms most other machine-learning approaches across most cancer types. Compared to traditional machine-learning-based patient classifiers, netDx results are more interpretable, visualizing the decision boundary in the context of patient similarity space. When patient similarity is defined by pathway-level gene expression, netDx identifies biological pathways important for outcome prediction, as demonstrated in breast cancer and asthma. netDx can serve as a patient classifier and as a tool for discovery of biological features characteristic of disease. We provide a free software implementation of netDx with automation workflows.

Publication rates of abstracts presented at major interventional radiology conferences.

PURPOSE: We aimed to determine the publication rate and factors predictive of publication of oral presentations at the annual meetings of the Cardiovascular and Interventional Radiology Society of Europe (CIRSE) and the Society of Interventional Radiology (SIR). METHODS: Keywords and authors from oral presentation abstracts at the 2012 CIRSE and SIR annual meetings were used to search PubMed and GoogleScholar for subsequent publication. Logistic regression was performed to identify whether number of authors, country of origin, subject category, methodology, study type, and/or study results were predictive of publication. RESULTS: A total of 421 abstracts (CIRSE-126, SIR-295) met the inclusion criteria. The overall publication rate across both conferences was 44.9%. Time from conference presentation to publication was 15±8.9 months for CIRSE and 16.3±8.8 months for SIR (P > 0.05), with a combined time interval of 15.9±8.8 months for both. The median impact factor of published abstracts was 2.075 (interquartile range, 2.075-2.775) for CIRSE and 2.093 (2.075-2.856) for SIR (P > 0.05). The most common country of origin for published abstracts was Germany (27.1%) at CIRSE and the United States (69%) at SIR. Logistic regression did not identify factors that were predictive of future publication. CONCLUSION: Publication rates were similar for CIRSE and SIR. Factors such as country of origin, topic of study and study results were not predictive of future publication. Authors should not be discouraged from submitting their work to journals based on these factors.

DEVELOPING AN IPAD® APPLICATION FOR DATA COLLECTION IN A RHEUMATOLOGY RESEARCH CLINIC.

OBJECTIVES: Clinical research data are often collected on paper and later inputted onto an electronic database. This method is time consuming and potentially introduces errors. Therefore, to make primary data collection more efficient and less error prone we aimed to develop a touch-screen application for data collection in a psoriatic arthritis research clinic and compared it with the pre-existing paper-based system. METHODS: We developed a Web application using Java and optimized it for the iPad®. It highlights missing fields for physicians in real time, and only permits submission of data collection form after corrections are made. For its evaluation, seven physicians participated, and before each patient visit they were randomly assigned paper or iPad® data entry. Number of errors, length of visit, and time between clinic visit and completion of data entry were measured. RESULTS: A total of 106 patients seen in the clinic who agreed to participate were randomly assigned to be evaluated by clinic physicians using the iPad® (fifty-three patients) or a paper protocol (fifty-three patients). On average, 3.34 omissions were found per paper form, of which 2.24 would have been detected on the iPad®. The iPad® increased the mean patient encounter time from 37.2 minutes to 46.5 minutes, but eliminated delay between a clinic visit and its data entry. CONCLUSIONS: Entering data using the iPad® application makes the patient encounter slightly longer, but reduces "missing fields." It also eliminates the delay between clinic visit and data entry thus improving the efficiency of clinical data capture in a research setting.