We describe IBEX, an open-source tool designed for medical imaging radiomics studies, and how to use this tool. In addition, some published works that have used IBEX for uncertainty analysis and model building are showcased.
Imaging Biomarker Explorer (IBEX) is an open-source tool for medical imaging radiomics work. The purpose of this paper is to describe how to use IBEX’s graphical user interface (GUI) and to demonstrate how IBEX calculated features have been used in clinical studies. IBEX allows for the import of DICOM images with DICOM radiation therapy structure files or Pinnacle files. Once the images are imported, IBEX has tools within the Data Selection GUI to manipulate the viewing of the images, measure voxel values and distances, and create and edit contours. IBEX comes with 27 preprocessing and 132 feature choices to design feature sets. Each preprocessing and feature category has parameters that can be altered. The output from IBEX is a spreadsheet that contains: 1) each feature from the feature set calculated for each contour in a data set, 2) image information about each contour in a data set, and 3) a summary of the preprocessing and features used with their selected parameters. Features calculated from IBEX have been used in studies to test the variability of features under different imaging conditions and in survival models to improve current clinical models.
In medicine, patient disease diagnosis typically incorporates a large number of diagnostic exams such as x-rays, ultrasound, computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET) scans to assist in determining the course of patient care. While physicians use these images to qualitatively assess patient's diagnosis, there may be additional quantitative features that can be extracted to guide patient care. The rationale is that these features may represent proteomic and genomic patterns expressed on the macroscopic scale1. Combining this quantitative information with the current clinical information, e.g., patient demographics, may allow more individualized patient care. This is the theory behind radiomics: feature analysis of images on a voxel level. The features typically fall into 5 main categories: gray level co-occurrence matrix, gray level run length matrix, neighborhood intensity difference matrix, histogram, and shape.
Imaging Biomarker Explorer (IBEX) is an open-source tool for radiomics work2. The graphical user interface (GUI) was developed at MD Anderson Cancer Center with the goal of facilitating the extraction and calculation of quantitative features to assist in decision making in cancer care. A source code3 and a stand-alone4 version are available online. IBEX calculates the 5 most common categories of features used in medical radiomics with parameters that can be set for each feature category. The categories are: gray level co-occurrence matrix5, gray level run length matrix6,7, intensity, neighborhood intensity difference matrix8, and shape. Since IBEX is open source, it allows for harmonized feature extraction results across institutions to easily compare different radiomics studies. All features within IBEX are described in the initial paper by Zhang et al.2
The purpose of this manuscript is to provide guidance on how to use IBEX and to demonstrate its applications through peer-reviewed published studies from the MD Anderson radiomics group. Since its release to the public in 2015, IBEX has been used to calculate features from CT, PET, and MRI scan images by the MD Anderson radiomics group, typically investigating features to improve clinical survival models9,10,11,12,13,14,15,16,17,18,19,20 and by outside institutions21,22,23,24. Additional guidance on software tools that can be used for the steps in radiomics research that are not included in IBEX can be found in Court et al.25
A general introduction to the workflow of IBEX will help to organize data properly before starting radiomics projects utilizing IBEX. If importing DICOM images, IBEX requires that each patient have their own folder with their DICOM images. DICOM radiation structure set is optional to include in the patient folder, but is recommended instead of using the contouring platform in IBEX. To assist with importing all patients for a specific study, all patient folders can be placed into one folder together so that all data may be imported into IBEX using only one step. If importing patients from Pinnacle, it is best to have the structure set with the patient plan. As patients may have multiple image sets and plans within Pinnacle, it is best to know which image set and plan are correct before importing. If computation time is a concern, reducing the number of image slices for a patient can drastically reduce time. For example, if only the liver is of interest in a study but the patients have full body CT scans, reducing the DICOM slices to only the extent of the area of interest can shorten computation time (e.g., reducing the DICOM from 300 slices to 50 slices can take 1/6th the time). There are different tools available to perform this slice reduction, from manual to semi-automatic.
1. Install IBEX
NOTE: To install a source-code version go to step 1.1. Alternatively, to install a stand-alone version go to step 1.2.
2. Set the Location
Note: The images are imported to this set location and the data for this study are stored here as well. IBEX uses the most previously applied location as the default location when re-launched.
3. Import Images
4. Viewing Images and Regions of Interest (ROIs)
5. Edit ROI
6. Contours in the Data Set
7. Create Feature Set
8. Output Features
9. Statistical Model Building
The output from IBEX is a spreadsheet (see Figure 4) that contains 3 tabs. The "Results" tab contains the feature values for each ROI in the data set (Figure 4A). The "Data Info." tab contains information about the images taken from each ROI in the data set (Figure 4B). The "Feature Info." tab contains a comprehensive list of features used with the parameters selected for the feature category and the preprocessing used for that category of features (Figure 4C).
The IBEX calculated features from medical images have been utilized in several contexts. Hunter et al. used an early version of IBEX to identify robust image features19. Fave et al. investigated the uncertainty in radiomics features of 4D CT thoracic scans collected at different respiratory phases, peak tube voltages, and tube currents9. This study found intra-patient variation to be less than inter-patient variation for most features when tube voltage and current were varied, making these factors negligible. The reproducibility of the feature from cone beam CT (CBCT) images was then evaluated using IBEX10. In this study, features calculated from lung CBCT images were found to be reproducible when the same protocol and manufacturer were used only when breathing motion was small. Image preprocessing impact on feature values was subsequently evaluated. The study showed that 39 of the 55 features studied had at least one preprocessing technique that resulted in significant stratification for overall survival using Cox proportional hazards models indicating that different preprocessing may be needed for each feature11. The uncertainty in features from perfusion CT images has also been evaluated using IBEX. Yang et al. showed that radiomics features were not dependent on time between contrast administration and CT scan, and that 86.9% of features were reproducible with an inter-session stability concordance correlations coefficient greater than or equal to 0.916. Lastly, a phantom was designed to test inter-scanner variability on a subset of features15. Texture strength was found to be the most consistent feature while busyness was found to vary the most.
The radiomics features from IBEX are also often used for model building, typically looking at overall survival, local-regional control, and freedom from distant metastases. Fried et al. identified 8 radiomics features from non-small cell lung cancer (NSCLC) patients' CT scans that when implemented into a Cox proportional hazards models for overall survival, loco-regional control, and distant metastases significantly improved Kaplan-Meier stratification when compared to models that only used clinical data20. Similarly, Fave et al. found radiomics features that improved patient stratification in survival curves12. Their study used weekly CT images and calculated changes in the lung radiomics features. Calculated features in the model had four different pre-processing methods: (1) thresholding, (2) threshold and bit depth, (3) thresholding and smoothing, (4) thresholding, bit depth, and smoothing; and the best pre-processing method was chosen for each feature individually before being tested in the Cox proportional hazards models. Hunter et al. also showed that radiomics features can predict tumor shrinking in NSCLC patients while exploring different thresholds and bit depth rescale values18.
PET image's radiomics features and their prognostic value have also been investigated using IBEX. Fried et al. scaled standardized uptake values (SUVs) by rounding SUVs to the nearest whole number and then subtracting the minimum SUV for that ROI from the rest13. Energy and solidity were found to statistically improve an overall survival model when included compared to the model when only conventional clinical factors were included. These two radiomics features were also found to be able to identify subgroups of patients who received a benefit or detriment from dose escalation14. Similarly, van Rossum et al. found an increase in the c-index for a clinical prediction model of pathologic complete response when including radiomics features into clinical models17.
Figure 1: IBEX main home page. Main page for IBEX with icons for each section. Each of these sections are described in Sections 2 – 6. Please click here to view a larger version of this figure.
Figure 2: Data selection window. GUI window is used for data selection manipulation. The window comes with buttons to alter the appearance of the images as described in steps 4.4 – 4.10. Please click here to view a larger version of this figure.
Figure 3: ROI editor window. GUI window is used for ROI manipulation. The window comes with the same buttons to alter the appearance as in the data selection as well as buttons to alter the ROIs. The ROI manipulation is described in Section 5. Please click here to view a larger version of this figure.
Figure 4: IBEX results worksheet. IBEX outputs three pages of information in a worksheet. The first page (A) contains the feature values for each ROI, the second page (B) contains information about the images that the ROIs were drawn on, and the third page (C) contains information about the features and preprocessing used. The outputs for this figure are from a phantom study where features were calculated using Butterworth smoothing and 8-bit depth rescaling, shown in panel C, column E, rows 5 and 6. Please click here to view a larger version of this figure.
IBEX is a powerful tool for medical imaging radiomics research. It has thus far mostly been used for radiation oncology purposes in studies conducted by the MD Anderson radiomics group. IBEX allows for manipulation of ROIs and calculation of features within 5 main feature categories. The source code version of IBEX allows the user to design applications that are not already part of IBEX, such as gray level zone matrix features.
The main steps involved in IBEX are the import of images, contouring of ROIs, selection of ROIs for data set, and creation of feature set. Accurate contours are necessary as features are only calculated within these areas and thus inaccurate contours will provide inaccurate feature values. Thus, any relationship found between these features calculated on inaccurate contours and outcomes will be spurious. Parameter selection for the features is also a vital step. For example, changing the step size for the gray level co-occurrence matrix can impact the features calculated from the matrix. This could depend on image type (i.e., MRI, CT, or PET), site for investigation (e.g., NSCLC vs. head and neck), and the purpose of the study (e.g., creating survival models vs. linking image features with genomics). Feature parameters should be selected based on a physical or biological reasoning, e.g., is there a reason that a step size of 4 would be biologically relevant in a co-occurrence matrix? Feature parameters can also be selected based on previous studies that have found certain feature parameters to correlate with outcomes or biological expressions. IBEX has 27 preprocessing modules and 132 features available for selection, along with allowing altering of parameters for each preprocessing module and feature category, which makes it an adaptable tool for many types of radiomics studies.
There are several general limitations in radiomics research that apply when using any software. For example, image features have been shown to depend on image acquisition parameters such as voxel size and scanner15,27. A limitation of all software is that there are many parameters that can be altered for each feature and the default values may not be appropriate for the specific study. Users must be vigilant, and research previously used parameter settings for similar studies and evaluate the applicability of the settings. The quality of the contours and the inherent inter- and intra-observer variability can also affect the calculation of features. Owens et al. showed that calculated features are more robust when using auto-contouring tools28. The features calculated for radiomics studies are human-engineered features and may not fully convey the features observed by the visual perception system. In addition, these features may be highly correlated to one another creating difficulties when analyzing the results. One specific limitation of IBEX is that the current version lacks the availability to calculate wavelet features; however, our group intends to include these features in future versions.
There are several alternative software platforms available for the calculation of image features25. Some advantages of IBEX include the fact that it is freely available, is well documented2, and allows users detailed control of the image feature calculations. IBEX also displays the processed patient images (e.g., after smoothing), so the user can visualize the impact of any image processing – this is useful, for example, when confirming that the preprocessing has not over-smoothed the images. Similarly, IBEX can export the actual co-occurrence matrix and intensity histograms; this can be useful when delving deeper into the image features.
IBEX has been solely used for cancer studies, mostly focused on radiation therapy. However, future studies can branch out to other cancer therapies or even outside of the cancer field. For example, Kassner et al. used radiomics to predict hemorrhagic transformation in patients with acute ischemic stroke29. IBEX could also be used in radiomics studies of this type.
IBEX also has continual maintenance. For example, a study by Fave et al. found that 5 features (busyness, coarseness, gray level non-uniformity, run length non-uniformity, and energy) were strongly volume dependent and corrected their formulas11. These updated formulas have been included in the updated release of IBEX. Additionally, there is a google group30 that has users post questions that other users then answer. This continual improvement of IBEX in addition to the current capabilities of IBEX and its availability make it a prime source for radiomics studies.
The authors have nothing to disclose.
Rachel Ger is funded by the Rosalie B. Hite Graduate Fellowship and American Legion Auxiliary Fellowship. Carlos Cardenas has been funded by the George M. Stancel PhD Fellowship in the Biomedical Sciences. The development of IBEX was funded by the NCI (R03 CA178495).
Excel | Microsoft Office | Any version of excel should work. | |
Matlab | MathWorks | Only use IBEX on 32 bit Matlab 2011a or 64 bit Matlab 2014b. |