Cancer imaging phenomics toolkit: quantitative imaging analytics for precision diagnostics and predictive modeling of clinical outcome

Abstract

The growth of multiparametric imaging protocols has paved the way for quantitative imaging phenotypes that predict treatment response and clinical outcome, reflect underlying cancer molecular characteristics and spatiotemporal heterogeneity, and can guide personalized treatment planning. This growth has underlined the need for efficient quantitative analytics to derive high-dimensional imaging signatures of diagnostic and predictive value in this emerging era of integrated precision diagnostics. This paper presents cancer imaging phenomics toolkit (CaPTk), a new and dynamically growing software platform for analysis of radiographic images of cancer, currently focusing on brain, breast, and lung cancer. CaPTk leverages the value of quantitative imaging analytics along with machine learning to derive phenotypic imaging signatures, based on two-level functionality. First, image analysis algorithms are used to extract comprehensive panels of diverse and complementary features, such as multiparametric intensity histogram distributions, texture, shape, kinetics, connectomics, and spatial patterns. At the second level, these quantitative imaging signatures are fed into multivariate machine learning models to produce diagnostic, prognostic, and predictive biomarkers. Results from clinical studies in three areas are shown: (i) computational neuro-oncology of brain gliomas for precision diagnostics, prediction of outcome, and treatment planning; (ii) prediction of treatment response for breast and lung cancer, and (iii) risk assessment for breast cancer.

Keywords: cancer imaging phenomics; open source software; precision diagnostics; radiogenomics; radiomics; treatment response.

Fig. 1

Overview of CaPTk’s functions: at the first level, CaPTk provides image preprocessing and feature extraction functions that can be used to generate an extensive QIP panel of features capturing various aspects of imaging signals, ranging from segmentation of tumors and its partitions, to extraction of textural and perfusion dynamic features, to population-wide spatial patterns of cancer, and fiber tracts. At the second level, these QIP features and maps are integrated into algorithmically complex diagnostic and predictive models, aiming to achieve precision diagnosis and guidance of treatment, prediction of clinical outcome, and estimation of molecular characteristics of tumors.

Fig. 2

Various aspects of the diffusion based surgical planning tools: (a) tracking through edema made possible with multicompartment modeling of diffusion data; (b) atlas-based reconstruction of tracts, resilient to mass effect induced tract distortions, and vulnerability map of the brain indicating the global effect of the resection and treatment; and (c) the surgical plan with the tumor and surrounding eloquent tract.

Fig. 3

An overview of the CaPTk software architecture. Command-line and GUI of CaPTk communicate with individual applications for preprocessing, basic analysis, and decisions support outcomes via function calls (black arrows). Applications may be tightly integrated in CaPTk, accessed as C++ objects via a documented API, or applications may be external software, such as Confetti, launched via system calls. Integrated applications utilize low-level libraries, such as ITK or libraries developed specifically for CaPTk for common tasks. Data are passed between libraries and returned to integrated applications in the form of ITK and OpenCV data structures (green arrow). Results are presented graphically through the GUI (light blue arrow) or saved to disk (red arrows). External applications return data directly to disk storage. The documented APIs allow applications written by external users of CaPTk to access algorithms at each level of the toolkit through function calls and ITK image passing.

Fig. 4

Example of (a) parenchymal complexity feature extraction and (b) breast cancer case-control classification in conjunction based on breast parenchymal density (PD) alone compared to the CaPTk texture feature extraction panel.

Fig. 5

Kaplan–Meier survival curves for the replication cohort. Actual survival on $x$-axis is compared among each of the three survival groups based on predictions generated by the SPI. med, medium SPI HR, Hazard ratio.

Fig. 6

Survival curves as function of (a) FTV, after first visit during neoadjuvant chemotherapy and (b) Jacobian heterogeneity when FTV is greater than the mean value (between images before and first visits).

Fig. 7

Survival analysis of two clusters of the early-stage NSCLC patients with respect to (a) death and (b) nodal failure.

Fig. 8

(a) Sagittal postgadolinum T1-weighted images with recurrence probability maps on preop scan, calculated via CaPTk. (b) Actual recurrence scan, about 3 months later.

Fig. 9

Examples of how QIP features are integrated into imaging signatures of molecular characteristics of glioblastoma. (a–c) Distributions of the PHI by EGFRvIII expression status. Statistical significance was evaluated via a two-tailed paired $t$-test comparing between the two distributions in the (a) discovery, (b) replication, and (c) combined cohorts. (d) ROC curves of four-way classification of glioblastoma into its molecular subtypes, using extensive radiogenomic signatures synthesized using machine learning.

Fig. 10

Intrinsic imaging phenotypes of breast cancer tumors via unsupervised clustering of multiparametric MRI features. The columns represent tumors and the rows features, showing four distinct phenotypes, related to tumor gene expression and hormone (ER/PR) receptor status.