Histogram analysis of diffusion-weighted magnetic resonance imaging as a biomarker to predict LNM in T3 stage rectal carcinoma

Yang Zhou, Rui Yang, Yuan Wang, Meng Zhou, Xueyan Zhou, JiQing Xing, Xinxin Wang, Chunhui ZhangBMC Medical Imaging 2021 · DOI

TL;DR

This study leverages histogram analysis of multi-parametric MRI (T2WI, DWI, ADC, T2-map) from primary rectal tumors to predict lymph node metastasis (LNM) in T3 stage rectal cancer. By combining these imaging features with clinical data, they developed a nomogram that significantly improves LNM prediction accuracy compared to traditional methods, offering a more precise tool for treatment planning.

What Problem Does This Paper Address?

Accurately determining lymph node metastasis (LNM) status in rectal cancer is critical for guiding treatment decisions, particularly for locally advanced (T3) tumors. While MRI is the standard for staging, its ability to reliably detect LNM is insufficient, often leading to either overtreatment (unnecessary neoadjuvant chemoradiotherapy) or undertreatment. Current MRI assessment relies on morphological criteria of lymph nodes, which can be unreliable, especially for smaller nodes. This paper aims to improve LNM prediction by exploring quantitative histogram features derived from multi-parametric MRI of the primary tumor itself, hypothesizing that tumor heterogeneity, as captured by these features, can serve as a biomarker for LNM.

Methods

The researchers conducted a retrospective analysis of 175 patients with stage T3 rectal cancer who underwent preoperative multi-parametric MRI. They extracted histogram features from regions of interest (ROIs) manually delineated on T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), ADC maps, and T2 maps. These histogram features quantify the distribution of signal intensities within the tumor.

Key methodological steps included:

  • Multi-parametric MRI Acquisition: Standard T2WI and DWI sequences were acquired. T2 maps and ADC maps were derived from DWI.
  • ROI Delineation: ROIs were manually drawn on the maximum cross-sections of the primary tumor on T2WI and DWI images. These ROIs were then mapped to the ADC and T2-map images.
  • Histogram Feature Extraction: For each ROI, various histogram features were calculated, including mean, median, standard deviation, coefficient of variation (CV), kurtosis, skewness, and mode.
  • Statistical Analysis:
    • Univariate and Multivariate Logistic Regression: Used to identify independent predictors of LNM from clinical, imaging, and histogram features.
  • Receiver Operating Characteristic (ROC) Curve Analysis: Employed to evaluate the diagnostic performance (sensitivity, specificity, AUC) of individual predictors and a combined model.
  • Nomogram Construction: A clinical-imaging-histogram nomogram was developed to provide a visual and quantitative tool for predicting LNM risk.
  • Calibration Plots and Decision Curve Analysis: Used to assess the nomogram's accuracy and clinical utility.

Datasets

The study utilized a dataset of 175 patients with stage T3 rectal cancer who underwent total mesorectal resection and preoperative MRI between January 2015 and May 2017. The data included clinical information, preoperative MRI images (T2WI, DWI, ADC, T2-map), and postoperative pathological reports for LNM status. No new datasets were introduced.

Key Insights and Ideas

  • Tumor Heterogeneity as a Biomarker: The core idea is that the internal heterogeneity of the primary rectal tumor, as quantified by histogram features, is linked to its propensity to metastasize to lymph nodes. This moves beyond solely assessing lymph node morphology.
  • Multi-parametric MRI Synergy: Combining features from T2WI, DWI, and T2-maps provides a more comprehensive picture of tumor characteristics than any single sequence alone.
  • Specific Histogram Features Matter: Certain histogram features, particularly from DWI (CV, Mode) and T2WI (Kurtosis), along with clinical factors (CA199, mrN stage), emerged as significant independent predictors of LNM.
  • Nomogram for Practical Application: The development of a clinical-imaging-histogram nomogram offers a user-friendly tool for clinicians to estimate LNM risk, potentially guiding treatment decisions more effectively.

Key Results

  • Independent Predictors: Stepwise logistic regression identified CA199 level, mrN stage, T2WI Kurtosis, DWI CV, DWI Mode, and T2-map P5 as independent predictors of LNM.
  • Combined Model Performance: The combined model (clinical + imaging histogram features) achieved an Area Under the ROC Curve (AUC) of 0.860, with a sensitivity of 72.8% and a specificity of 85.5%. This is a significant improvement over previous studies focusing on single sequences.
  • Nomogram Utility: The developed nomogram demonstrated good discrimination and calibration, with decision curve analysis suggesting a favorable potential clinical effect.

Limitations and Future Work

  • Retrospective Single-Center Study: The study's retrospective nature and single-center design introduce potential biases in patient selection and data acquisition.
  • Limited Texture Analysis: While histogram analysis was performed, other texture analysis techniques (e.g., gray-level co-occurrence matrix) were not explored.
  • Small Sample Size: The authors acknowledge that the universal applicability of the nomogram may be limited due to the relatively small sample size.

Future Work: The authors suggest that larger, multi-center studies are needed to validate these findings and the developed nomogram, potentially incorporating a wider range of texture features to further enhance predictive accuracy.

Summary updated

BibTeX
@article{zhou2021histogram,
  author = {Yang Zhou and Rui Yang and Yuan Wang and Meng Zhou and Xueyan Zhou and JiQing Xing and Xinxin Wang and Chunhui Zhang},
  title = {{Histogram analysis of diffusion-weighted magnetic resonance imaging as a biomarker to predict LNM in T3 stage rectal carcinoma}},
  journal = {BMC Medical Imaging},
  year = {2021},
  doi = {10.1186/s12880-021-00706-0},
}
Figure