Radiomics Driven Diffusion Weighted Imaging Sensing Strategies for Zone-Level Prostate Cancer Sensing

Chris Dulhanty, Linda Wang, Maria Cheng, Hayden Gunraj, Farzad Khalvati, Masoom A. Haider, Alexander Wong

TL;DR

This paper introduces novel radiomics-driven strategies (Zone-X and Zone-DR) for more precise, zone-level prostate cancer detection using Diffusion Weighted Imaging (DWI). These methods significantly outperform traditional clinical heuristics, offering a promising avenue for improved cancer sensing and personalized treatment planning.

What Problem Does This Paper Address?

Prostate cancer is a common malignancy, but early and accurate detection is crucial for effective treatment and improved patient outcomes. While Diffusion Weighted Imaging (DWI) has emerged as a promising non-invasive tool for cancer sensing, current clinical practice relies on relatively simple "clinical heuristics" (threshold-based approaches) that may not fully capture the complex spatial and textural characteristics of cancerous tissue. Furthermore, these methods often operate at a coarser level, whereas understanding cancer at a more granular "zone-level" within the prostate could lead to more targeted biopsies and treatments. This paper aims to bridge this gap by developing advanced radiomics-driven strategies that leverage DWI data for more accurate and localized prostate cancer sensing.

Methods

The authors propose two main categories of radiomics-driven sensing strategies:

  • Zone-X Sensing Strategies: These are based on hand-engineered radiomic features. The study utilizes a modified radiomic sequence from existing literature, incorporating first-order statistics, Haralick features, Kirsch features, and Gabor features. These features are then fed into standard machine learning classifiers: Support Vector Machines (SVM), Logistic Regression (LR), and Random Forest (RF). These Zone-X strategies are novel in their application to zone-level prostate cancer sensing using DWI.
  • Zone-DR (Discovery Radiomics) Sensing Strategy: This is a more advanced, two-part approach.
  • Machine-Driven Sequencer Design Discovery: This involves using a deep convolutional neural network (CNN) architecture to automatically discover optimal radiomic sequencer designs tailored for zone-level sensing. The goal is to find architectures that maximize performance (measured by AUC).
  • Data-Driven Sequencer Parameter Discovery: Once the sequencer architecture is designed, this step uses an iterative optimization method (Adam optimizer) to learn the specific parameters of the discovered sequencers, optimizing them for accurate cancer grading using a cross-entropy loss function. This "discovery radiomics" approach aims to uncover radiomic features directly from the data, rather than relying on predefined ones.

The study also evaluates a Clinical Heuristics Driven Strategy as a baseline, using established thresholds for ADC (<1000 s/mm²) and CHB-DWI (>1000 s/mm²) to classify zones as cancerous.

Datasets

The study utilizes a dataset comprising Diffusion Weighted Imaging (DWI) data from 101 patients. This includes both Apparent Diffusion Coefficient (ADC) imaging and Computed High-b Value Diffusion Weighted Imaging (CHB-DWI).

  • Data Acquisition: Images were acquired using a Philips Achieva 3.0 T MRI machine.
  • Annotation: Prostate glands were manually segmented into 10 anatomical zones by an expert radiologist. Each zone was assigned a PI-RADS score.
  • Pathological Verification: 72 patients with PI-RADS scores of 3 or above underwent biopsies. Pathological assessment using the Gleason grading system provided ground truth for cancer presence and grade.
  • Dataset Size: A total of 12,466 prostate zones were extracted, with 12,361 being negative-grade and 135 being positive-grade (pathology-verified).
  • Splitting and Validation: The dataset was split into five stratified folds at the patient level to ensure representation of positive-grade zones in each fold. A 5-fold cross-validation approach was used for evaluation.

Key Insights and Ideas

  • Zone-Level Sensing is Crucial: The study strongly suggests that moving beyond slice-level analysis to zone-level prostate cancer sensing is beneficial, as different zones can exhibit distinct characteristics and cancer behavior.
  • Radiomics Outperforms Heuristics: Both hand-engineered (Zone-X) and discovery-driven (Zone-DR) radiomics strategies significantly outperform traditional clinical heuristics in terms of Area Under the Curve (AUC), indicating a more robust ability to distinguish cancerous from non-cancerous tissue.
  • Discovery Radiomics (Zone-DR) Shows Promise: The Zone-DR approach, which learns radiomic features directly from data, generally achieved competitive or superior performance compared to Zone-X strategies, especially in terms of sensitivity and specificity.
  • Modality Choice Matters: The choice between ADC and CHB-DWI sensing modalities can be optimized based on clinical needs. ADC appears to offer higher sensitivity (useful for screening to minimize missed cancers), while CHB-DWI provides higher specificity (useful for surgical planning to minimize false positives).
  • Explainability is Key: The Zone-DR strategy's ability to visualize critical regions using explainability methods (like GSInquire) provides valuable insights into why the model makes certain predictions, aligning with radiologists' visual cues (e.g., signal hyperintensity in CHB-DWI). This transparency is crucial for clinical adoption.
  • CHB-DWI Offers Superior Characterization: CHB-DWI generally provides better cancer tissue characterization than ADC, leading to higher accuracy and a better balance of sensitivity and specificity when used with radiomics strategies.

Key Results

  • AUC Improvement: Zone-X and Zone-DR strategies achieved higher AUC scores than clinical heuristics. For example, Zone-DR achieved an average AUC of ~0.86 for ADC, compared to ~0.79 for clinical heuristics.
  • Highest Sensitivity/Specificity:
    • ADC: Zone-DR and Zone-SVM achieved the highest sensitivity and specificity, respectively.
  • CHB-DWI: Zone-DR and Zone-LR achieved the highest sensitivities, while Zone-DR, Zone-LR, and Zone-SVM achieved the highest specificities.
  • Trade-off Optimization: The study demonstrates that the balance between sensitivity and specificity can be tuned, allowing for strategies tailored to specific clinical scenarios (e.g., screening vs. surgical planning).
  • Visualizations: Figure 7 shows that Zone-DR's critical regions for positive cancer zones align with signal hyperintensity in CHB-DWI, a feature radiologists look for. For negative zones, Zone-DR identified larger regions, suggesting a broader contextual analysis.

Limitations and Future Work

  • Dataset Size: While 101 patients and 12,466 zones are substantial, larger and more diverse datasets could further validate the findings and improve generalizability.
  • Clinical Heuristics Baseline: The specific clinical heuristics used as a baseline are based on prior studies; exploring variations or more contemporary heuristic approaches could provide a more nuanced comparison.
  • Explainability Method: While GSInquire is presented as a state-of-the-art method, further exploration of different explainability techniques could offer complementary insights.
  • Future Work: The authors suggest that these radiomics-driven DWI sensing strategies hold significant potential for aiding radiologists in zone-level prostate cancer screening and could be a powerful tool for improving detection and treatment planning. Further research could focus on integrating these methods into clinical workflows and validating them in prospective studies.

Summary updated

Abstract

Prostate cancer is the most commonly diagnosed cancer in North American men; however, prognosis is relatively good given early diagnosis. This motivates the need for fast and reliable prostate cancer sensing. Diffusion weighted imaging (DWI) has gained traction in recent years as a fast non-invasive approach to cancer sensing. The most commonly used DWI sensing modality currently is apparent diffusion coefficient (ADC) imaging, with the recently introduced computed high-b value diffusion weighted imaging (CHB-DWI) showing considerable promise for cancer sensing. In this study, we investigate the efficacy of ADC and CHB-DWI sensing modalities when applied to zone-level prostate cancer sensing by introducing several radiomics driven zone-level prostate cancer sensing strategies geared around hand-engineered radiomic sequences from DWI sensing (which we term as Zone-X sensing strategies). Furthermore, we also propose Zone-DR, a discovery radiomics approach based on zone-level deep radiomic sequencer discovery that discover radiomic sequences directly for radiomics driven sensing. Experimental results using 12,466 pathology-verified zones obtained through the different DWI sensing modalities of 101 patients showed that: (i) the introduced Zone-X and Zone-DR radiomics driven sensing strategies significantly outperformed the traditional clinical heuristics driven strategy in terms of AUC, (ii) the introduced Zone-DR and Zone-SVM strategies achieved the highest sensitivity and specificity, respectively for ADC amongst the tested radiomics driven strategies, (iii) the introduced Zone-DR and Zone-LR strategies achieved the highest sensitivities for CHB-DWI amongst the tested radiomics driven strategies, and (iv) the introduced Zone-DR, Zone-LR, and Zone-SVM strategies achieved the highest specificities for CHB-DWI amongst the tested radiomics driven strategies. Furthermore, the results showed that the trade-off between sensitivity and specificity can be optimized based on the particular clinical scenario we wish to employ radiomic driven DWI prostate cancer sensing strategies for, such as clinical screening versus surgical planning. Finally, we investigate the critical regions within sensing data that led to a given radiomic sequence generated by a Zone-DR sequencer using an explainability method to get a deeper understanding on the biomarkers important for zone-level cancer sensing.
BibTeX
@inproceedings{dulhanty2020radiomics,
  author = {Chris Dulhanty and Linda Wang and Maria Cheng and Hayden Gunraj and Farzad Khalvati and Masoom A. Haider and Alexander Wong},
  title = {{Radiomics Driven Diffusion Weighted Imaging Sensing Strategies for Zone-Level Prostate Cancer Sensing}},
  booktitle = {Sensors},
  year = {2020},
  url = {https://openreview.net/forum?id=RTyWtB582G},
}
Figure