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.
BACKGROUND AND PURPOSE: Neuronal ceroid lipofuscinoses are a group of neurodegenerative disorders characterized by the accumulation of autofluorescent lipopigments in neuronal cells. As a result of storage material in the brain and retina, clinical manifestations include speech delay, cognitive dysfunction, motor regression, epilepsy, vision loss, and early death. At present, 14 different ceroid lipofuscinosis (CLN) genes are known. Recently, the FDA approved the use of recombinant human proenzyme of tripeptidyl-peptidase 1 for CLN2 disease, while phase I/IIa clinical trials for gene therapy in CLN3 and CLN6 are ongoing. Early diagnosis is, therefore, key to initiating treatment and arresting disease progression. Neuroimaging features of CLN1, CLN2, CLN3, and CLN5 diseases are well-described, with sparse literature on other subtypes. We aimed to investigate and expand the MR imaging features of genetically proved neuronal ceroid lipofuscinoses subtypes at our institution and also to report the time interval between the age of disease onset and the diagnosis of neuronal ceroid lipofuscinoses. MATERIALS AND METHODS:We investigated and analyzed the age of disease onset and neuroimaging findings (signal intensity in periventricular, deep, and subcortical white matter, thalami, basal ganglia, posterior limb of the internal capsule, insular/subinsular regions, and ventral pons; and the presence or absence of supratentorial and/or infratentorial atrophy) of patients with genetically proved neuronal ceroid lipofuscinoses at our institution. This group consisted of 24 patients who underwent 40 brain MR imaging investigations between 1993 and 2019, with a male preponderance (male/female ratio ¼ 15:9). RESULTS:The mean ages of disease onset, first brain MR imaging, and diagnosis of neuronal ceroid lipofuscinoses were 4.70 6 3.48 years, 6.76 6 4.49 years, and 7.27 6 4.78 years, respectively. Findings on initial brain MR imaging included T2/FLAIR hypointensity in the thalami (n ¼ 22); T2/FLAIR hyperintensity in the periventricular and deep white matter (n ¼ 22), posterior limb of the internal capsule (n ¼ 22), ventral pons (n ¼ 19), and insular/subinsular region (n ¼ 18); supratentorial (n ¼ 21) and infratentorial atrophy (n ¼ 20). Eight of 9 patients who had follow-up neuroimaging showed progressive changes. CONCLUSIONS:We identified reported classic neuroimaging features in all except 1 patient with neuronal ceroid lipofuscinoses in our study. CLN2, CLN5, and CLN7 diseases showed predominant cerebellar-over-cerebral atrophy. We demonstrate that abnormal signal intensity in the deep white matter, posterior limb of the internal capsule, and ventral pons is more common than previously reported in the literature. We report abnormal signal intensity in the insular/subinsular region for the first time. The difference in the median time from disease onset and diagnosis was 1.5 years.ABBREVIATIONS: CLN ¼ ceroid lipofuscinosis; DWM ¼ deep white matter; IQR ¼ interquartile range; I-SI ¼ insular/subinsular region; NCL ¼ neuronal ceroi...
BackgroundCurrently there is no consensus agreement on the degree of enhancement in normal temporomandibular joints (TMJ) in children, which makes it difficult for clinicians to distinguish between the presence/absence of mild synovitis. Quantitative measurements of synovial and condylar enhancement may be useful additions to current qualitative methods on early MRI diagnosis and follow up of TMJ involvement in JIA. The purpose of the study is to establish thresholds/tendencies for quantitative measures that enable distinction between mild TMJ involvement and normal TMJ appearance based on the degree of synovial and bone marrow enhancement in JIA patients.MethodsTMJ MRI examinations in 67 children with JIA and in 24 non-rheumatologic children who underwent MRI for neurologic/orbit indications were retrospectively assessed. As a priori determined TMJs of JIA patients were categorized into three groups by experienced staff radiologists based on the degree of synovial and condylar enhancement: no active disease (rheumatologic control), mild and moderate/severe findings. The signal intensity (SI) of the synovial tissue around each condyle and of the bone marrow was measured to calculate the enhancement ratio (ER) and relative SI change. The ER was calculated using signal to noise ratios, while relative SI change was calculated using signal intensities alone. Quantitative measurements of synovial and condylar enhancement of TMJs with mild or moderate/severe findings were compared with the rheumatologic and non-rheumatologic controls.ResultsMean ER values were significantly different between the TMJs without active disease and those with mild and moderate/severe synovial enhancement, with highest values in the moderate/severe group (P < 0.0001). Similar findings were seen for condylar enhancement with P < 0.005. Relative SI change was unable to differentiate TMJs with mild synovitis from the two controls (P > 0.10). 27/60 (45 %) TMJs without active disease had osteochondral changes. 8/40 (20 %) TMJs in the mild group did not demonstrate any synovial thickening.ConclusionsQuantitative signal to noise ratios of TMJ synovial and condylar enhancement generate thresholds/tendencies, which offer additional information to differentiate mild synovitis from normal TMJs in JIA patients. Osteochondral changes and synovial thickening may not be reliable indicators of active TMJ involvement and should be differentiated from synovial enhancement.
BACKGROUND AND PURPOSE: MRI is routinely performed following brain AVM after treatment in children. Our aim was to determine the predictive values of contrast-enhanced MR imaging and TOF-MRA for brain AVM recurrence in children, compared with conventional angiography and the role of 3D rotational angiography-MR imaging fusion in these cases. MATERIALS AND METHODS: We included all pediatric patients with brain AVMs during an 18-year period with angiographically documented obliteration after treatment. Patients underwent 3T MR imaging, including contrast-enhanced MR imaging, TOF-MRA, and conventional angiography, with a subset undergoing 3D rotational angiography. The predictive values of contrast-enhanced MR imaging and TOF-MRA for brain AVM recurrence were determined. CTA sections reconstructed from 3D rotational angiography were coregistered with and fused to 3D-T1WI for analysis.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.