A Comparative Study on the Potential of Unsupervised Deep Learning-based Feature Selection in Radiomics.

In Radiomics, deep learning-based systems for medical image analysis play an increasing role. However, due to the better explainability, feature-based systems are still preferred, especially by physicians. Often, high-dimensional data and low sample size pose different challenges (e.g. increased risk of overfitting) to machine learning systems. By removing irrelevant and redundant features from the data, feature selection is an effective way of pre-processing. The research in this study is focused on unsupervised deep learning-based methods for feature selection. Five recently proposed algorithms are compared regarding their applicability and efficiency on seven data sets in three different sample applications. It was found that deep learning-based feature selection leads to improved classification results compared to conventional methods, especially for small feature subsets. Clinical Relevance - The exploration of distinctive features and the ability to rank their importance without the need for outcome information is a potential field of application for unsupervised feature selection methods. Especially in multiparametric radiology, the number of features is increasing. The identification of new potential biomarkers is important both for treatment and prevention.

Effects of simulated dose variation on contrast-enhanced CT-based radiomic analysis for Non-Small Cell Lung Cancer.

PURPOSE: To examine the potential effect of CT dose variation on radiomic features in vivo using simulated contrast-enhanced CT dose reduction in patients with non-small lung cell cancer (NSCLC). METHODS: In this retrospective study, we included 69 patients (25 females, 44 males, median age 66 years) with histologically proven NSCLC who underwent a whole contrast-enhanced body FDG-PET/CT for primary staging. To simulate different CT dose levels, we used an algorithm to simulate low-dose CT images based on a noise model derived from phantom experiments. The tumor lesions and reference regions in the paraspinal muscle were manually segmented to obtain three-dimensional regions of interest. Radiomic feature extraction was performed using the PyRadiomics toolbox. The median relative feature value deviation was assessed for each feature and each dose level. RESULTS: The mean segmented tumor volume was 340 ml. T-stages of the primary tumors were primarily T3/4. For NSCLCs, the median relative feature value deviation in the lowest dose images varied for the calculated features from 52.2% to -49.5%. In general, dose-dependent deviations of feature values showed a monotonous increase or decrease with decreasing dose levels. Statistical analyses revealed significant differences between the dose levels in 91% of features. CONCLUSIONS: We examined the effects of simulated CT dose reduction on the values of radiomic features in primary NSCLC and showed significant deviations of varying degrees in numerous feature classes. This is a theoretical indicator of potential influence under real conditions, which should be taken into account in clinical use.

ImFEATbox: a toolbox for extraction and analysis of medical image features.

PURPOSE: In medical imaging, the digital post-processing and analysis of acquired images has become an important research field. Topics include various applications of image processing and machine learning aiming to assist radiologists in their diagnostic work. A crucial step in successfully implementing such systems is finding appropriate mathematical descriptions to reflect characteristics of acquired images. Which features are the most meaningful ones strongly depends on the underlying scientific/diagnostic question and the image itself. This makes researching, implementing and testing features time-consuming and cost-intensive. In our work, we aim to address this issue by creating ImFEATbox, a publicly available toolbox to extract and analyze image features for a wide range of applications. METHODS: To reduce the amount of time spent for choosing the right features, we provide an assortment of feature extraction algorithms which are suitable for a broad variety of medical image processing problems. The toolbox includes both global and local features as well as feature descriptors. While being primarily developed in MATLAB, the majority of our algorithms is also available in Python to enable access to a wider range of researchers. RESULTS: We tested the applicability of ImFEATbox on an FDG-PET/CT data set of 12 patients diagnosed with lung cancer and an MRI data set of 50 patients with prostate lesions. Employing the implemented algorithms in an exemplary manner, we are able to demonstrate its potential for different scientific problems, e.g., show differences between features, indicate redundancies in extracted feature sets by means of a correlation analysis and training a SVM to distinguish between high-risk and low-risk prostate lesions. CONCLUSION: ImFEATbox provides a variety of feature extraction algorithms suitable for a large number of post-processing and analysis applications in medical imaging. The toolbox is publicly available and can thus be beneficial to a wide range of researchers working on medical image analysis.

Automated reference-free detection of motion artifacts in magnetic resonance images.

OBJECTIVES: Our objectives were to provide an automated method for spatially resolved detection and quantification of motion artifacts in MR images of the head and abdomen as well as a quality control of the trained architecture. MATERIALS AND METHODS: T1-weighted MR images of the head and the upper abdomen were acquired in 16 healthy volunteers under rest and under motion. Images were divided into overlapping patches of different sizes achieving spatial separation. Using these patches as input data, a convolutional neural network (CNN) was trained to derive probability maps for the presence of motion artifacts. A deep visualization offers a human-interpretable quality control of the trained CNN. Results were visually assessed on probability maps and as classification accuracy on a per-patch, per-slice and per-volunteer basis. RESULTS: On visual assessment, a clear difference of probability maps was observed between data sets with and without motion. The overall accuracy of motion detection on a per-patch/per-volunteer basis reached 97%/100% in the head and 75%/100% in the abdomen, respectively. CONCLUSION: Automated detection of motion artifacts in MRI is feasible with good accuracy in the head and abdomen. The proposed method provides quantification and localization of artifacts as well as a visualization of the learned content. It may be extended to other anatomic areas and used for quality assurance of MR images.