RESUMEN
Millions of people around the world suffer from Parkinson's disease, a neurodegenerative disorder with no remedy. Currently, the best response to interventions is achieved when the disease is diagnosed at an early stage. Supervised machine learning models are a common approach to assist early diagnosis from clinical data, but their performance is highly dependent on available example data and selected input features. In this study, we explore 23 single photon emission computed tomography (SPECT) image features for the early diagnosis of Parkinson's disease on 646 subjects. We achieve 94 % balanced classification accuracy in independent test data using the full feature space and show that matching accuracy can be achieved with only eight features, including original features introduced in this study. All the presented features can be generated using a routinely available clinical software and are therefore straightforward to extract and apply.
Asunto(s)
Enfermedad de Parkinson , Diagnóstico Precoz , Humanos , Aprendizaje Automático , Enfermedad de Parkinson/diagnóstico por imagen , Tomografía Computarizada de Emisión de Fotón ÚnicoRESUMEN
A heart-to-mediastinum (H/M) ratio of 1.6 or greater on planar 123I-iobenguane (123I-MIBG) images identifies heart failure patients at low risk of experiencing an adverse cardiac event. This phase-4 study used standardized phantoms to assess the intercamera, intracamera, and interhead variability in H/M ratio determinations from planar cardiac 123I-MIBG imaging using commercially available, dual-head γ-cameras. Methods: A fillable thorax phantom was developed to simulate the typical uptake of 123I-MIBG. The phantom had a nominal H/M ratio of 1.6 on the reference camera. Commercial cameras used in the study were dual-head and capable of 90° configuration for cardiac imaging. The target sample size was 8 units (examples) per camera model. Two imaging technologists independently analyzed planar images of simulated 123I-MIBG uptake from the thorax phantom. H/M was the ratio of the average counts per pixel of the heart and mediastinum regions of interest. The primary endpoint, intercamera variability in H/M ratio from head 1, was determined for each camera model via comparison with the H/M ratio on the reference camera. Only cameras with at least 8 units tested (n ≥ 8) were included in the primary analysis. Intracamera and interhead variability in the H/M ratio were also evaluated. Results: Nine camera models were studied. The mean H/M ratio ranged from 1.342 to 1.677. The primary analysis (6 camera models) using a mixed-model, repeated-measures analysis showed no significant difference in H/M ratio between any camera model and the reference camera. Intracamera variability (head 1) in the H/M ratio among camera models with 8 units or more was high, with SDs ranging from 0.0455 to 0.1193. Interhead variability was low (SDs of the interhead difference, 0.017-0.074). Conclusion: Commonly used γ-cameras produced H/M ratios from simulated 123I-MIBG phantom images that were not significantly different from those on the reference camera. This finding indicates that the results of previous clinical trials of 123I-MIBG, involving many different clinical sites and camera models, are valid. The assessment of the performance of a given camera unit using an 123I planar phantom before H/M results from 123I-MIBG imaging are used for classifying risk in heart failure patients is encouraged.