Heterogeneous Image-based Classification Using Distributional Data Analysis

TL;DR

A novel imaging-based distributional data analysis (DDA) approach that incorporates the probability (quantile) distribution of the pixel-level features as covariates in a scalar-on-functional quantile regression model for risk prediction in Hepatocellular carcinoma.

Abstract

Diagnostic imaging has gained prominence as potential biomarkers for early detection and diagnosis in a diverse array of disorders including cancer. However, existing methods routinely face challenges arising from various factors such as image heterogeneity. We develop a novel imaging-based distributional data analysis (DDA) approach that incorporates the probability (quantile) distribution of the pixel-level features as covariates. The proposed approach uses a smoothed quantile distribution (via a suitable basis representation) as functional predictors in a scalar-on-functional quantile regression model. Some distinctive features of the proposed approach include the ability to: (i) account for heterogeneity within the image; (ii) incorporate granular information spanning the entire distribution; and (iii) tackle variability in image sizes for unregistered images in cancer applications. Our primary goal is risk prediction in Hepatocellular carcinoma that is achieved via predicting the change in tumor grades at post-diagnostic visits using pre-diagnostic enhancement pattern mapping (EPM) images of the liver. Along the way, the proposed DDA approach is also used for case versus control diagnosis and risk stratification objectives. Our analysis reveals that when coupled with global structural radiomics features derived from the corresponding T1-MRI scans, the proposed smoothed quantile distributions derived from EPM images showed considerable improvements in sensitivity and comparable specificity in contrast to classification based on routinely used summary measures that do not account for image heterogeneity. Given that there are limited predictive modeling approaches based on heterogeneous images in cancer, the proposed method is expected to provide considerable advantages in image-based early detection and risk prediction.

Figures (2)

• Figure 1: A schematic of the proposed analytical pipeline. The leftmost panel captures the 5 different phases of the T1-MRI liver scan that are all registered to a common template for a given subject. Subsequently the EPM images are computed from these T1-MRI scans, and then empirical quantile distributions followed by smoothed quantile distributions (based on quantlet basis expansions) were computed. The arterial phase of T1-MRI is used for segmentation of tumor and also used to compute the structural radiomics features. These two types of features, along with demographics are used for three classification tasks that include early detection, risk stratification, and diagnosis (case vs controls).
• Figure 2: Visualizations of lesion EPM images with segmentations for lesion grades 2-5 and healthy control. Liver masks are shown in yellow; lesion masks in red and surrounding peri-lesional areas are shaded in blue. Mean empirical quantiles are shown along with 95% confidence bands for each lesion grade and controls. For cases, quantiles are shown across all lesional (red) and peri-lesional (blue) areas, and for controls, quantiles are shown across all liver masks (yellow). Distributions of log-lesion volumes are also shown for each lesion grade. Observed lesion volumes ranged from 0.17-0.950 cm$^3$, with approximate diameters of 0.70-12.2 cm.