RESUMEN
PURPOSE: The purpose of this study was to create an algorithm to detect and classify pulmonary nodules in two categories based on their volume greater than 100 mm3 or not, using machine learning and deep learning techniques. MATERIALS AND METHOD: The dataset used to train the model was provided by the organization team of the SFR (French Radiological Society) Data Challenge 2019. An asynchronous and parallel 3-stages pipeline was developed to process all the data (a data "pre-processing" stage; a "nodule detection" stage; a "classifier" stage). Lung segmentation was achieved using 3D U-NET algorithm; nodule detection was done using 3D Retina-UNET and classifier stage with a support vector machine algorithm on selected features. Performances were assessed using area under receiver operating characteristics curve (AUROC). RESULTS: The pipeline showed good performance for pathological nodule detection and patient diagnosis. With the preparation dataset, an AUROC of 0.9058 (95% confidence interval [CI]: 0.8746-0.9362) was obtained, 87% yielding accuracy (95% CI: 84.83%-91.03%) for the "nodule detection" stage, corresponding to 86% specificity (95% CI: 82%-92%) and 89% sensitivity (95% CI: 84.83%-91.03%). CONCLUSION: A fully functional pipeline using 3D U-NET, 3D Retina-UNET and classifier stage with a support vector machine algorithm was developed, resulting in high capabilities for pulmonary nodule classification.
Asunto(s)
Inteligencia Artificial , Neoplasias Pulmonares , Nódulos Pulmonares Múltiples , Aprendizaje Profundo , Humanos , Neoplasias Pulmonares/clasificación , Neoplasias Pulmonares/diagnóstico por imagen , Nódulos Pulmonares Múltiples/clasificación , Nódulos Pulmonares Múltiples/diagnóstico por imagen , Tomografía Computarizada por Rayos XRESUMEN
Volume is a basic physical property of cells; however, it has been poorly investigated in cell biology so far, mostly because it is difficult to measure it precisely. Recently, large efforts were made to experimentally measure mammalian cell size and used mass, density, or volume as proxies for cell size. Here, we describe a method enabling cell volume measurements for single living cells. The method is based on the principle of fluorescent dye exclusion and can be easily implemented in cell biology laboratories. As this method is very versatile, it can be used for cells of different sizes, adherent or growing in suspension, over several cell cycles and is independent of cell shape changes. The method is also compatible with traditional cell biology tools such as epifluorescence imaging or drug treatments.
Asunto(s)
Tamaño de la Célula , Rastreo Celular/métodos , Análisis de la Célula Individual/métodos , Ciclo Celular/genética , Forma de la Célula/genética , Colorantes Fluorescentes/químicaRESUMEN
Localization of single molecules in microscopy images is a key step in quantitative single particle data analysis. Among them, single molecule based super-resolution optical microscopy techniques require high localization accuracy as well as computation of large data sets in the order of 10(5) single molecule detections to reconstruct a single image. We hereby present an algorithm based on image wavelet segmentation and single particle centroid determination, and compare its performance with the commonly used gaussian fitting of the point spread function. We performed realistic simulations at different signal-to-noise ratios and particle densities and show that the calculation time using the wavelet approach can be more than one order of magnitude faster than that of gaussian fitting without a significant degradation of the localization accuracy, from 1 nm to 4 nm in our range of study. We propose a simulation-based estimate of the resolution of an experimental single molecule acquisition.