Purpose: The required training sample size for a particular machine learning (ML) model applied to medical imaging data is often unknown. The purpose of this study was to provide a descriptive review of current sample-size determination methodologies in ML applied to medical imaging and to propose recommendations for future work in the field. Methods: We conducted a systematic literature search of articles using Medline and Embase with keywords including “machine learning,” “image,” and “sample size.” The search included articles published between 1946 and 2018. Data regarding the ML task, sample size, and train-test pipeline were collected. Results: A total of 167 articles were identified, of which 22 were included for qualitative analysis. There were only 4 studies that discussed sample-size determination methodologies, and 18 that tested the effect of sample size on model performance as part of an exploratory analysis. The observed methods could be categorized as pre hoc model-based approaches, which relied on features of the algorithm, or post hoc curve-fitting approaches requiring empirical testing to model and extrapolate algorithm performance as a function of sample size. Between studies, we observed great variability in performance testing procedures used for curve-fitting, model assessment methods, and reporting of confidence in sample sizes. Conclusions: Our study highlights the scarcity of research in training set size determination methodologies applied to ML in medical imaging, emphasizes the need to standardize current reporting practices, and guides future work in development and streamlining of pre hoc and post hoc sample size approaches.

Objective: To study the influence of ethnicity on the risk of developing juvenile idiopathic arthritis (JIA) in a multiethnic community of patients with unrestricted access to health care.Methods: A questionnaire on ethnicity was distributed to all patients with JIA being followed up at the Hospital for Sick Children in Toronto, Ontario, Canada. Of 1,082 patients, 859 (79.4%) responded to the questionnaire. To calculate the relative risk (RR) of developing JIA in this study cohort, the results were compared with data from the age‐matched general population of the Toronto metropolitan area (TMA) as provided in the 2001 census from Statistics Canada.Results: European descent was reported by 69.7% of the patients with JIA compared with a frequency of 54.7% in the TMA general population, whereas a statistically significantly lower than expected percentage of the patients with JIA reported having black, Asian, or Indian subcontinent origin. Children of European origin had a higher RR for developing any of the JIA subtypes except polyarticular rheumatoid factor (RF)–positive JIA, and were particularly more likely to develop the extended oligoarticular and psoriatic subtypes. A higher frequency of enthesitis‐related JIA was observed among patients of Asian origin, while those of black origin or native North American origin were more likely to develop polyarticular RF‐positive JIA.Conclusion: In this multiethnic cohort, European descent was associated with a significantly increased risk of developing JIA, and the distribution of JIA subtypes differed significantly across ethnic groups.

Cryopyrin-associated periodic syndrome (CAPS) is a rare, heterogeneous disease entity associated with NLRP3 gene mutations and increased interleukin-1 (IL-1) secretion. Early diagnosis and rapid initiation of IL-1 inhibition prevent organ damage. The aim of the study was to develop and validate diagnostic criteria for CAPS. An innovative process was followed including interdisciplinary team building, item generation: review of CAPS registries, systematic literature review, expert surveys, consensus conferences for item refinement, item reduction and weighting using 1000Minds decision software. Resulting CAPS criteria were tested in large cohorts of CAPS cases and controls using correspondence analysis. Diagnostic models were explored using sensitivity analyses. The international team included 16 experts. Systematic literature and registry review identified 33 CAPS-typical items; the consensus conferences reduced these to 14. 1000Minds exercises ranked variables based on importance for the diagnosis. Correspondence analysis determined variables consistently associated with the diagnosis of CAPS using 284 cases and 837 controls. Seven variables were significantly associated with CAPS (p<0.001). The best diagnosis model included: Raised inflammatory markers (C-reactive protein/serum amyloid A) plus ≥two of six CAPS-typical symptoms: urticaria-like rash, cold-triggered episodes, sensorineural hearing loss, musculoskeletal symptoms, chronic aseptic meningitis and skeletal abnormalities. Sensitivity was 81%, specificity 94%. It performed well for all CAPS subtypes and regardless of NLRP3 mutation. The novel approach integrated traditional methods of evidence synthesis with expert consensus, web-based decision tools and innovative statistical methods and may serve as model for other rare diseases. These criteria will enable a rapid diagnosis for children and adults with CAPS.

Objective: Primary angiitis of the central nervous system (PACNS) is a severe and ill‐defined neurologic disease. The goal of this study was to characterize the presenting features, treatment, and neurologic outcome of PACNS in children (cPACNS) and to define the predictors of disease progression in order to identify high‐risk patients with cPACNS.Methods: The cohort comprised consecutive patients diagnosed as having cPACNS based on clinical and vascular imaging findings, including identification of arterial stenosis on conventional angiography or magnetic resonance (MR) angiography. Disease progression was defined angiographically at >3 months after initial angiography. Clinical data obtained in prospectively collected standardized assessments and results of laboratory tests, including detection of cerebrospinal fluid abnormalities, were noted, and neuroimaging studies were reanalyzed. Predictors of progression were identified and tested in multivariate regression models.Results: Sixty‐two consecutive patients with cPACNS (38 male, 24 female, median age 7.2 years) were included. Two distinct subgroups were identified, those with progressive disease and those with nonprogressive disease. Progressive cPACNS was found in 20 of 62 children and was predicted by a clinical presentation of neurocognitive dysfunction, multifocal parenchymal lesions on MR imaging, and evidence of distal stenoses on angiography.Conclusion: The spectrum of PACNS in children includes both progressive and nonprogressive forms. Characteristic features at diagnosis can be used to predict later progression, to identify a distinct high‐risk cPACNS cohort, and to help guide selection of patients for immunosuppressive therapy.

Takayasu arteritis is a devastating vasculitis of the aorta and its major branches. The clinical manifestations in paediatric patients are less specific than in adults: in children the disease presents with fever, arthralgias and hypertension. Intramural inflammation results in narrowing of the blood vessel lumen and therefore hypoperfusion of the parenchyma. Conventional angiography is the gold standard diagnostic procedure. Corticosteroids, cyclophosphamide, MTX and biological therapies such as TNF-α blocking agents are treatment options.