Dataset Information

Deep Feature Stability Analysis Using CT Images of a Physical Phantom Across Scanner Manufacturers, Cartridges, Pixel Sizes, and Slice Thickness.

ABSTRACT: Image acquisition parameters for computed tomography scans such as slice thickness and field of view may vary depending on tumor size and site. Recent studies have shown that some radiomics features were dependent on voxel size (= pixel size × slice thickness), and with proper normalization, this voxel size dependency could be reduced. Deep features from a convolutional neural network (CNN) have shown great promise in characterizing cancers. However, how do these deep features vary with changes in imaging acquisition parameters? To analyze the variability of deep features, a physical radiomics phantom with 10 different material cartridges was scanned on 8 different scanners. We assessed scans from 3 different cartridges (rubber, dense cork, and normal cork). Deep features from the penultimate layer of the CNN before (pre-rectified linear unit) and after (post-rectified linear unit) applying the rectified linear unit activation function were extracted from a pre-trained CNN using transfer learning. We studied both the interscanner and intrascanner dependency of deep features and also the deep features' dependency over the 3 cartridges. We found some deep features were dependent on pixel size and that, with appropriate normalization, this dependency could be reduced. False discovery rate was applied for multiple comparisons, to mitigate potentially optimistic results. We also used stable deep features for prognostic analysis on 1 non-small cell lung cancer data set.

SUBMITTER: Paul R

PROVIDER: S-EPMC7289258 | biostudies-literature | 2020 Jun

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Deep Feature Stability Analysis Using CT Images of a Physical Phantom Across Scanner Manufacturers, Cartridges, Pixel Sizes, and Slice Thickness.

Paul Rahul R Shafiq-Ul Hassan Mohammed M Moros Eduardo G EG Gillies Robert J RJ Hall Lawrence O LO Goldgof Dmitry B DB

Tomography (Ann Arbor, Mich.) 20200601 2

Image acquisition parameters for computed tomography scans such as slice thickness and field of view may vary depending on tumor size and site. Recent studies have shown that some radiomics features were dependent on voxel size (= pixel size × slice thickness), and with proper normalization, this voxel size dependency could be reduced. Deep features from a convolutional neural network (CNN) have shown great promise in characterizing cancers. However, how do these deep features vary with changes ...[more]

PMID: 32548303

Similar Datasets

Project description:Computed Tomography (CT) imaging of the lung, reported in Hounsfield Units (HU), can be parameterized as a quantitative image biomarker for the diagnosis and monitoring of lung density changes due to emphysema, a type of chronic obstructive pulmonary disease (COPD). CT lung density metrics are global measurements based on lung CT number histograms, and are typically a quantity specifying either the percentage of voxels with CT numbers below a threshold, or a single CT number below which a fixed relative lung volume, nth percentile, falls. To reduce variability in the density metrics specified by CT attenuation, the Quantitative Imaging Biomarkers Alliance (QIBA) Lung Density Committee has organized efforts to conduct phantom studies in a variety of scanner models to establish a baseline for assessing the variations in patient studies that can be attributed to scanner calibration and measurement uncertainty.Data were obtained from a phantom study on CT scanners from four manufacturers with several protocols at various tube potential voltage (kVp) and exposure settings. Free from biological variation, these phantom studies provide an assessment of the accuracy and precision of the density metrics across platforms solely due to machine calibration and uncertainty of the reference materials. The phantom used in this study has three foam density references in the lung density region, which, after calibration against a suite of Standard Reference Materials (SRM) foams with certified physical density, establishes a HU-electron density relationship for each machine-protocol. We devised a 5-step calibration procedure combined with a simplified physical model that enabled the standardization of the CT numbers reported across a total of 22 scanner-protocol settings to a single energy (chosen at 80 keV). A standard deviation was calculated for overall CT numbers for each density, as well as by scanner and other variables, as a measure of the variability, before and after the standardization. In addition, a linear mixed-effects model was used to assess the heterogeneity across scanners, and the 95% confidence interval of the mean CT number was evaluated before and after the standardization.We show that after applying the standardization procedures to the phantom data, the instrumental reproducibility of the CT density measurement of the reference foams improved by more than 65%, as measured by the standard deviation of the overall mean CT number. Using the lung foam that did not participate in the calibration as a test case, a mixed effects model analysis shows that the 95% confidence intervals are [-862.0 HU, -851.3 HU] before standardization, and [-859.0 HU, -853.7 HU] after standardization to 80 keV. This is in general agreement with the expected CT number value at 80 keV of -855.9 HU with 95% CI of [-857.4 HU, -854.5 HU] based on the calibration and the uncertainty in the SRM certified density.This study provides a quantitative assessment of the variations expected in CT lung density measures attributed to non-biological sources such as scanner calibration and scanner x-ray spectrum and filtration. By removing scanner-protocol dependence from the measured CT numbers, higher accuracy and reproducibility of quantitative CT measures were attainable. The standardization procedures developed in study may be explored for possible application in CT lung density clinical data.

Project description:BackgroundNucleus is a fundamental task in microscopy image analysis and supports many other quantitative studies such as object counting, segmentation, tracking, etc. Deep neural networks are emerging as a powerful tool for biomedical image computing; in particular, convolutional neural networks have been widely applied to nucleus/cell detection in microscopy images. However, almost all models are tailored for specific datasets and their applicability to other microscopy image data remains unknown. Some existing studies casually learn and evaluate deep neural networks on multiple microscopy datasets, but there are still several critical, open questions to be addressed.ResultsWe analyze the applicability of deep models specifically for nucleus detection across a wide variety of microscopy image data. More specifically, we present a fully convolutional network-based regression model and extensively evaluate it on large-scale digital pathology and microscopy image datasets, which consist of 23 organs (or cancer diseases) and come from multiple institutions. We demonstrate that for a specific target dataset, training with images from the same types of organs might be usually necessary for nucleus detection. Although the images can be visually similar due to the same staining technique and imaging protocol, deep models learned with images from different organs might not deliver desirable results and would require model fine-tuning to be on a par with those trained with target data. We also observe that training with a mixture of target and other/non-target data does not always mean a higher accuracy of nucleus detection, and it might require proper data manipulation during model training to achieve good performance.ConclusionsWe conduct a systematic case study on deep models for nucleus detection in a wide variety of microscopy images, aiming to address several important but previously understudied questions. We present and extensively evaluate an end-to-end, pixel-to-pixel fully convolutional regression network and report a few significant findings, some of which might have not been reported in previous studies. The model performance analysis and observations would be helpful to nucleus detection in microscopy images.

Dataset Information

Deep Feature Stability Analysis Using CT Images of a Physical Phantom Across Scanner Manufacturers, Cartridges, Pixel Sizes, and Slice Thickness.

Publications

Deep Feature Stability Analysis Using CT Images of a Physical Phantom Across Scanner Manufacturers, Cartridges, Pixel Sizes, and Slice Thickness.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets