Cambridge recently reported the first study using quantitative image analysis of cellular heterogeneity in breast tumours to complement genomic profiling. In order to reliably identify multi-parametric imaging signatures, which can be associated with clearly characterised genetic and micro-environmental subtypes, we must address significant technical problems associated with the accuracy and precision of calculated imaging parameters in individual pixels and the selection of heterogeneity analysis metrics.
Many image-derived biomarkers show highly variable accuracy and precision in areas with different proportional tissue constituents, signal-to-noise ratios and/or significant motion. Many imaging studies employ summary statistics for large regions of interest (ROI) to minimise the contribution from individual voxels in which estimates of individual biomarkers may be highly inaccurate.
Manchester has developed heterogeneity metrics that clearly outperform conventional summary metrics in a number of areas involving diagnosis, tumour grading and treatment response. Current work now focuses on developing magnetic resonance (MR) imaging heterogeneity metrics in preclinical models.
We will develop this theme by addressing major challenges restricting robust techniques for the quantification of tumour heterogeneity, focussing on both development of robust imaging approaches and thorough biological qualification. We intend to develop the potential of imaging-based heterogeneity measures using both MR and positron emission tomography (PET).
In order to improve the quality of reconstructed PET images, we will develop and evaluate advanced spatiotemporal models to describe radiotracer distribution that are specifically designed for application in oncology. The particular challenges associated with this are the estimation of an input function and the effects of hypoxia and diffusion on tracer distribution.
We will develop, validate and qualify statistically robust techniques for identifying tissue subtypes across individuals using multimodality data, and qualify these tissue signatures using immunohistochemical and genetic analyses. Initial studies to statistically combine data will focus on existing brain data sets with anatomical, diffusion weighted and dynamic contrast MR. The approach will then be extended to encompass PET data, evaluated in tissues affected by physiological motion. Neuro-oncological MR data will enable us to validate and test our methods.
We will integrate serial clinical imaging and biopsies (including liquid biopsies) with the aim of detecting and unravelling response/resistance induced changes to both novel agents and repurposed drugs. Immunohistochemical and genetic analysis - whole genome sequencing, whole exome sequencing and RNA sequencing - of biopsies will be used to qualify imaging biomarker based classifications. This will allow development of optimised and reduced clinical imaging protocols. The multi-dimensional data (genomic, automated microscopy-image analysis and radiological images) will be integrated by developing a robust integrated database and analysis pipeline with proper visualisation tools to give a systems view of cancer at diagnosis and as it evolves in response to treatment.
Heterogeneity in drug delivery represents an important mechanism of treatment resistance. We will focus initially on preclinical and clinical studies in the brain to establish analytical and imaging approaches to allow investigation of drug delivery and the interactions with radiotherapy and molecularly targeted agents. Depending on results, we anticipate the development of clinical studies in primary lung and cerebral metastatic disease.
A study in breast cancer will use clips placed during surgery to correlate 3D imaging data – including MR and PET - with histological maps of hypoxia, angiogenesis and proliferation. The use of different multiple signals should allow a closer approximation to the “ground truth” of the current biological status in different regions of a tumour.