{"doi":"10.1016/j.ebiom.2021.103517","title":"Can deep learning classify stroke subtypes from chest X-rays?","abstract":"Deep neural networks have been shown to diagnose and predict risk of disease based on medical imaging [[1]Erickson BJ Korfiatis P Akkus Z Kline TL. Machine learning for medical imaging.Radiographics. 2017; 37: 505-515Crossref PubMed Scopus (484) Google Scholar]. Chest radiographs (x-ray or CXR) present a tremendous opportunity for deep learning algorithms. They are one of the most common tests in medicine and can be a window into systemic health and disease, especially for cardiovascular and respiratory systems. Most work in this field has been focused on using deep learning to mimic a radiologist's read of the CXR, akin to an automated written report [[2]Calli E Sogancioglu E van Ginneken B van Leeuwen KG Murphy K. Deep learning for chest X-ray analysis: a survey.Med Image Anal. 2021; 72102125Summary Full Text Full Text PDF PubMed Scopus (9) Google Scholar]. More recently, researchers have explored the use of deep learning to accomplish objectives not currently performed by radiologists such as risk estimation [[3]Lu MT Ivanov A Mayrhofer T Hosny A Aerts H Hoffmann U. Deep learning to assess long-term mortality from chest radiographs.JAMA Netw Open. 2019; 2e197416Crossref PubMed Scopus (32) Google Scholar,[4]Raghu VK Weiss J Hoffmann U Aerts HJWL Lu MT. Deep learning to estimate biological age from chest radiographs.JACC. 2021; Google Scholar] or complex diagnostic tasks [[2]Calli E Sogancioglu E van Ginneken B van Leeuwen KG Murphy K. Deep learning for chest X-ray analysis: a survey.Med Image Anal. 2021; 72102125Summary Full Text Full Text PDF PubMed Scopus (9) Google Scholar]. In this article, Jeong and colleagues explore another interesting classification problem by testing whether a deep learning model (they call ASTRO-X) can classify cardioembolic from noncardioembolic stroke based on a single CXR image [[5]Jeong H-G Kim BJ Kim T et al.Classification of cardioembolic stroke based on a deep neural network using chest radiographs.EBioMedicine. 2021; 69https://doi.org/10.1016/j.ebiom.2021.103466Summary Full Text Full Text PDF PubMed Scopus (1) Google Scholar]. Cardioembolic strokes account for 14–30% of all cerebral infarcts, and their incidence is rising, possibly due to increased life expectancy and better prevention of other causes of stroke (e.g., improved treatment for hypertension and dyslipidemia) [[6]Arboix A Alio J. Cardioembolic stroke: clinical features, specific cardiac disorders and prognosis.Curr Cardiol Rev. 2010; 6: 150-161Crossref PubMed Scopus (198) Google Scholar]. Cases of cardioembolic stroke are characterized by a major cardiac source of embolism without significant arterial disease [[6]Arboix A Alio J. Cardioembolic stroke: clinical features, specific cardiac disorders and prognosis.Curr Cardiol Rev. 2010; 6: 150-161Crossref PubMed Scopus (198) Google Scholar]. Diagnosing cardioembolic stroke involves a combination of neuroimaging, cardiac imaging, and laboratory measures. For example, the TOAST (Trial of Org 10172 in Acute Stroke Treatment) system, which was used as the gold standard in this study, states that a diagnosis of cardioembolic stroke requires identifying at least one cardiac source of an embolus and eliminating the possibility of large-artery atherosclerosis as the cause of thrombosis/embolism [[7]Adams Jr., HP Bendixen BH Kappelle LJ et al.Classification of subtype of acute ischemic stroke. Definitions for use in a multicenter clinical trial. TOAST. Trial of Org 10172 in Acute Stroke Treatment.Stroke. 1993; 24: 35-41Crossref PubMed Google Scholar]. In this work, the authors developed and tuned the ASTRO-X model using CXRs from 3,255 patients with acute ischemic stroke from a single institution. Patients with unknown or undetermined etiology of stroke were excluded (N = 2,175), and so the ASTRO-X model focuses on distinguishing cardioembolic from non-cardioembolic in those with known acute ischemic stroke. The final ASTRO-X model was evaluated in two testing sets: 809 different patients f","journal":"EBioMedicine","year":2021,"id":225944,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":0,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9582,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2021-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":282485,"name":"Michael T. Lu","orcid":"0000-0003-4696-9610","position":1,"is_corresponding":false},{"id":402710,"name":"Vineet K. Raghu","orcid":"0000-0003-3524-3945","position":0,"is_corresponding":true}],"reference_count":9,"raw_metadata":null,"created_at":"2026-07-18T23:54:30.292454Z","pmid":"34364166","pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}