Currently, magnetic resonance imaging is the most sensitive imaging technique for detecting cancerous processes in early stages. As for breast cancer, due to the tubular structure of the tissue, being formed by ducts, anisotropic diffusion should be considered instead of the general isotropic diffusion. Anisotropic diffusion is studied by applying a technique called Diffusion Tensor Imaging (DTI), where the diffusion gradient is applied by changing the magnetic field in several spatial directions.To date, the application of Multivariate Curve Resolution (MCR) models in diffusion sequences has demonstrated its ability to develop cancer biomarkers of easy clinical interpretation in the case of isotropic tissues, such as the prostate. But so far, it has never been applied in the case of anisotropic tissues, as the breast.Therefore, the main objective of this work is to obtain easy-to-interpret imaging biomarkers useful for early breast cancer diagnosis from diffusion magnetic resonance imaging based on the Diffusion Tensor using multivariate curve resolution (MCR) models. A classification model to identify healthy and tumor affected pixels is also proposed.

Functional MRI is, currently, the most sensitive technique in breast cancer for detecting early tumors, and perfusion (DCE-MRI) has become the most important sequence to depict and characterize angiogenesis and neovascularization. In this work, we propose the use of new biomarkers that are related to clear physiological phenomena, obtained from MCR-ALS as an alternative to curve-based pseudo-biomarkers and pharmacokinetics models. In order to provide a discrimination and prediction model between healthy tissue and cancer, we propose using PLS-DA with double cross-validation (2CV) and variable selection, repeated several times and obtaining excellent average results for the performance indexes (f-score: 0.9149, MCC: 0.8538, AUROC: 0.8794). After selecting the optimal prediction model, a unique probabilistic map called “virtual biopsy” that shows in different colors the probability that each pixel of the image has a tumor behavior is obtained, helping the specialist with the identification and characterization of breast tumors with only one easy-to-interpret biomarker map.

Currently, magnetic resonance imaging is the most sensitive imaging technique for detecting cancer processes in early stages. Regarding breast cancer, due to the characteristics of the tissue as it is formed by ducts (tubular structure), anisotropic diffusion should be considered instead of general isotropic Diffusion Weighted Imaging (DWI). Anisotropic diffusion is studied by applying a technique called Diffusion Tensor Imaging (DTI), where the diffusion gradient is applied with several different directions, calculated by Ordinary Least Squares (OLS) in clinical practice. In this paper, we propose a new DTI calculation method based on Partial Least Squares (PLS), which has some advantages over the traditional OLS calculation: i) the PLS model provides valid biomarkers (non-negative eigenvalues) in a larger percentage of pixels, improving the traditional OLS calculation and reducing the effect of noisier images; ii) OLS tensors are calculated pixel-by-pixel, whereas the PLS method calculates only one model taking advantage of the correlation structure between pixels with similar characteristics, obtaining more reliable estimations; iii) PLS performance is quite reliable when lowering the number of directions of the magnetic field, while this is not the case of OLS. PLS keeps providing a good solution even with low functional resolution equipment, reducing costs and acquisition times, which is an important advantage for its widespread use in value-based medicine-oriented clinical practice.

In current radiology practice, multi-parametric magnetic resonance imaging (mpMRI) has recently become a key tool in diagnostic and therapeutic decisions. Although it is based on the subjective assessment of T2-weighted images, as well as perfusion-weighted and diffusion-weighted sequences, further quantitative parameters can also be derived from them for improving lesion phenotyping. Despite these parameters are usually exploited in a univariate way, ignoring the benefits of a real multivariate approach, still it is the gold standard imaging technique to assess prostate cancer location and probability of malignancy. In this paper, pharmacokinetic (perfusion) and exponential (diffusion) clinical models, as well as latent variable-based multivariate statistical models like multivariate curve resolution-alternating least squares (MCR-ALS), have been calculated and analyzed with sequential multi block-partial least squares discriminant analysis (SMB-PLS-DA) including technique-block differentiation, in order to better assess for cancer aggressiveness based on Gleason scales. The best prediction result was achieved by the ordered combination of diffusion blocks (MCR-ALS and exponential models) and normalized T2 values. The perfusion blocks did not improve the results obtained by diffusion and T2-weighted based parameters alone, so they can be removed from the SMB-PLS-DA model.

Poster: ECR 2018 / C-0452 / Prediction of prostate tumor aggressiveness by DCE-MRI perfusion biomarkers and latent variables estimation methods by: E. Aguado-Sarrio, R. Sanz-Requena, L. Marti-Bonmati , J. M. Prats-Montalban, A. Ferrer-Riquelme; Valencia/ES

In this work, the capability of imaging biomarkers obtained from multivariate curve resolution-alternating least squares (MCR-ALS), in combination with those obtained from first and second-generation pharmacokinetic models, have been studied for improving prostate cancer tumor depiction using partial least squares-discriminant analysis (PLS-DA). The main goal of this work is to improve tissue classification properties selecting the best biomarkers in terms of prediction. A wrapped double cross-validation method has been applied for the variable selection process. Using the best PLS-DA model, prostate tissues can be classified obtaining 13.4% of false negatives and 7.4% of false positives. Using MCR-ALS biomarkers yields the best models in terms of parsimony and classification performance.