The Quantitative Feature Explore (QFExplore) plugin suite for the open-access cloud-based ePAD platform enables the exploration and validation of imaging biomarkers in a clinical environment. The latter include:
(i) the extraction, visualization and comparison of intensity- and texture-based quantitative imaging features (Fig. 1),
(ii) regional division of Regions Of Interests (ROI) to reveal tissue diversity (Fig. 2),
(iii) the construction, use and sharing of user-personalized statistical machine learning models (Fig. 3),
(iv) helper tools for image segmentation are also available (Fig. 4).
Figures illustrating the various plugins capabilities (i)-(iv) are shown below.
Imaging features available are:
- histogram bins of Pixel Intensity Distributions (PID),
- statistical moments of PIDs (i.e., mean, standard deviation, skewness, kurtosis),
- Gray-Level Co-occurrence Matrices (GLCMs),
- Riesz wavelets.
The machine learning model available is based on linear Support Vector Machines (SVMs).
A complete description of the plugin suite can be found in:
Roger Schaer, Yashin Dicente Cid, Emel Alkim, Sheryl John, Daniel L. Rubin and Adrien Depeursinge, Web-Based Tools for Exploring the Potential of Quantitative Imaging Biomarkers in Radiology: Intensity and Texture Analysis on the ePAD Platform, in: Biomedical Texture Analysis: Fundamentals, Applications and Tools, Elsevier, 2017.
Figure 1. (i) The user can compare the feature's values for various ROIs. GLCM contrast and correlation is higher for vascular ROIs.
Figure 2. (ii) Regional tissue diversity revealed by patch-based analysis.
Figure 3. (iii) Selection of labeled ROIs for training an SVM model to classify between normal and abnormal regions using Riesz features. The SVM model can be further shared with other colleagues.
Figure 4. (iii) Lung tissue classification of unseen ROIs using Riesz features (middle left, RieszOrder = 0, NumScales = 5), GLCMs features (middle right), Statistics features (bottom left) and Histogram features (bottom right, HistMin = 0, HistMax = 100, HistNumBins = 20, the bin values are used as features). Riesz and GLCMs are providing best results in this example.
Figure 5. (iv) Two images of automatically segmented lungs in a DICOM volume. The detected lung regions are highlighted with green image overlays in the front-end of ePAD.