Dataset Information

Convolution kernel and iterative reconstruction affect the diagnostic performance of radiomics and deep learning in lung adenocarcinoma pathological subtypes.

ABSTRACT:

Background

The aim of this study was to investigate the influence of convolution kernel and iterative reconstruction on the diagnostic performance of radiomics and deep learning (DL) in lung adenocarcinomas.

Methods

A total of 183 patients with 215 lung adenocarcinomas were included in this study. All CT imaging data was reconstructed with three reconstruction algorithms (ASiR at 0%, 30%, 60% strength), each with two convolution kernels (bone and standard). A total of 171 nodules were selected as the training-validation set, whereas 44 nodules were selected as the testing set. Logistic regression and a DL framework-DenseNets were selected to tackle the task. Three logical experiments were implemented to fully explore the influence of the studied parameters on the diagnostic performance. The receiver operating characteristic curve (ROC) was used to evaluate the performance of constructed models.

Results

In Experiments A and B, no statistically significant results were found in the radiomic method, whereas two and six pairs were statistically significant (P?ConclusionThe results demonstrated that DL was more susceptible to CT parameter variability than radiomics. Standard convolution kernel images seem to be more appropriate for imaging analysis. Further investigation with a larger sample size is needed.

SUBMITTER: Zhao W

PROVIDER: S-EPMC6775016 | biostudies-literature | 2019 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Convolution kernel and iterative reconstruction affect the diagnostic performance of radiomics and deep learning in lung adenocarcinoma pathological subtypes.

Zhao Wei W Zhang Wei W Sun Yingli Y Ye Yuxiang Y Yang Jiancheng J Chen Wufei W Gao Pan P Li Jianying J Li Cheng C Jin Liang L Wang Peijun P Hua Yanqing Y Li Ming M

Thoracic cancer 20190819 10

<h4>Background</h4>The aim of this study was to investigate the influence of convolution kernel and iterative reconstruction on the diagnostic performance of radiomics and deep learning (DL) in lung adenocarcinomas.<h4>Methods</h4>A total of 183 patients with 215 lung adenocarcinomas were included in this study. All CT imaging data was reconstructed with three reconstruction algorithms (ASiR at 0%, 30%, 60% strength), each with two convolution kernels (bone and standard). A total of 171 nodules ...[more]

PMID: 31426132

Similar Datasets

Project description:Several simplifications used in clinical implementations of the convolution∕superposition (C∕S) method, specifically, density scaling of water kernels for heterogeneous media and use of a single polyenergetic kernel, lead to dose calculation inaccuracies. Although these weaknesses of the C∕S method are known, it is not well known which of these simplifications has the largest effect on dose calculation accuracy in clinical situations. The purpose of this study was to generate and characterize high-resolution, polyenergetic, and material-specific energy deposition kernels (EDKs), as well as to investigate the dosimetric impact of implementing spatially variant polyenergetic and material-specific kernels in a collapsed cone C∕S algorithm.High-resolution, monoenergetic water EDKs and various material-specific EDKs were simulated using the EGSnrc Monte Carlo code. Polyenergetic kernels, reflecting the primary spectrum of a clinical 6 MV photon beam at different locations in a water phantom, were calculated for different depths, field sizes, and off-axis distances. To investigate the dosimetric impact of implementing spatially variant polyenergetic kernels, depth dose curves in water were calculated using two different implementations of the collapsed cone C∕S method. The first method uses a single polyenergetic kernel, while the second method fully takes into account spectral changes in the convolution calculation. To investigate the dosimetric impact of implementing material-specific kernels, depth dose curves were calculated for a simplified titanium implant geometry using both a traditional C∕S implementation that performs density scaling of water kernels and a novel implementation using material-specific kernels.For our high-resolution kernels, we found good agreement with the Mackie et al. kernels, with some differences near the interaction site for low photon energies (<500 keV). For our spatially variant polyenergetic kernels, we found that depth was the most dominant factor affecting the pattern of energy deposition; however, the effects of field size and off-axis distance were not negligible. For the material-specific kernels, we found that as the density of the material increased, more energy was deposited laterally by charged particles, as opposed to in the forward direction. Thus, density scaling of water kernels becomes a worse approximation as the density and the effective atomic number of the material differ more from water. Implementation of spatially variant, polyenergetic kernels increased the percent depth dose value at 25 cm depth by 2.1%-5.8% depending on the field size, while implementation of titanium kernels gave 4.9% higher dose upstream of the metal cavity (i.e., higher backscatter dose) and 8.2% lower dose downstream of the cavity.Of the various kernel refinements investigated, inclusion of depth-dependent and metal-specific kernels into the C∕S method has the greatest potential to improve dose calculation accuracy. Implementation of spatially variant polyenergetic kernels resulted in a harder depth dose curve and thus has the potential to affect beam modeling parameters obtained in the commissioning process. For metal implants, the C∕S algorithms generally underestimate the dose upstream and overestimate the dose downstream of the implant. Implementation of a metal-specific kernel mitigated both of these errors.

Project description:PURPOSE:Convolutional neural networks have become rapidly popular for image recognition and image analysis because of its powerful potential. In this paper, we developed a method for classifying subtypes of lung adenocarcinoma from pathological images using neural network whose that can evaluate phenotypic features from wider area to consider cellular distributions. METHODS:In order to recognize the types of tumors, we need not only to detail features of cells, but also to incorporate statistical distribution of the different types of cells. Variants of autoencoders as building blocks of pre-trained convolutional layers of neural networks are implemented. A sparse deep autoencoder which minimizes local information entropy on the encoding layer is then proposed and applied to images of size [Formula: see text]. We applied this model for feature extraction from pathological images of lung adenocarcinoma, which is comprised of three transcriptome subtypes previously defined by the Cancer Genome Atlas network. Since the tumor tissue is composed of heterogeneous cell populations, recognition of tumor transcriptome subtypes requires more information than local pattern of cells. The parameters extracted using this approach will then be used in multiple reduction stages to perform classification on larger images. RESULTS:We were able to demonstrate that these networks successfully recognize morphological features of lung adenocarcinoma. We also performed classification and reconstruction experiments to compare the outputs of the variants. The results showed that the larger input image that covers a certain area of the tissue is required to recognize transcriptome subtypes. The sparse autoencoder network with [Formula: see text] input provides a 98.9% classification accuracy. CONCLUSION:This study shows the potential of autoencoders as a feature extraction paradigm and paves the way for a whole slide image analysis tool to predict molecular subtypes of tumors from pathological features.

Dataset Information

Convolution kernel and iterative reconstruction affect the diagnostic performance of radiomics and deep learning in lung adenocarcinoma pathological subtypes.

Background

Methods

Results

Publications

Convolution kernel and iterative reconstruction affect the diagnostic performance of radiomics and deep learning in lung adenocarcinoma pathological subtypes.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets