Numerical observer for atherosclerotic plaque classification in spectral computed tomography

Auranuch Lorsakul, Georges El Fakhri, W. Worstell, Jinsong Ouyang, Yothin Rakvongthai, Andrew F. Laine, Quanzheng LiJournal of Medical Imaging 2016 · DOI

TL;DR

This paper introduces a novel numerical observer that leverages the full spectral information from multi-energy CT (MECT) to improve the classification of atherosclerotic plaques. By integrating spectral data across multiple energy bins, this observer significantly outperforms conventional and dual-energy CT methods in distinguishing plaque types, offering a more robust tool for assessing cardiovascular risk.

What Problem Does This Paper Address?

Atherosclerotic plaques are a major cause of cardiovascular events, and accurately characterizing them (e.g., distinguishing calcified from fatty plaques) is crucial for risk assessment. While spectral computed tomography (SCT) offers improved image quality and material differentiation compared to conventional CT, objective methods for evaluating its performance in plaque classification have been limited. Existing methods often struggle with the complex, multi-dimensional nature of SCT data, and there's a need for a robust observer model that can fully utilize the spectral information to assess diagnostic performance in a clinically relevant manner.

Methods

The authors developed a two-stage numerical observer.

  1. Preprocessing Stage: Each energy bin from the SCT data is processed using a channelized Hotelling observer (CHO). This involves decorrelating the spectral image within each bin using localized prewhitening and matched filtering with Laguerre-Gaussian channel functions. This stage extracts key features from each energy window.
  2. Integration Stage: The intermediate scores from all the CHOs (across all energy bins) are then integrated by a 1-D Hotelling observer. This observer uses an additional prewhitening and matched filter to combine the spectral information, yielding an overall figure of merit (FOM) such as signal-to-noise ratio (SNR) and area under the receiver operating characteristic curve (AUC).

The study also employed two task-based frameworks:

  • Signal-Known-Exactly/Background-Known-Exactly (SKE/BKE): A simplified model where signal properties are fully known.
  • Signal-Known-Statistically/Background-Known-Exactly (SKS/BKE): A more clinically realistic model that incorporates signal variability (e.g., in plaque size or density) to better mimic real-world scenarios.

Datasets

The study utilized simulated data generated from an extended cardiac-torso anthropomorphic digital phantom (XCAT). Three simulated plaque types were created: calcified plaque, fatty-mixed plaque, and iodine-mixed blood. These simulated plaques were inserted into a carotid artery model within the phantom. The data was reconstructed using a standard filtered backprojection (FBP) algorithm for various CT systems. No real patient data was used for training or evaluation.

Key Insights and Ideas

  • Full Spectral Utilization is Key: The core idea is that simply using two energy bins (as in DECT) is insufficient. By integrating information across multiple energy bins (as in MECT), a more comprehensive understanding of material properties can be achieved.
  • Clinical Realism Matters: The SKS/BKE framework, which accounts for signal variability, is crucial for accurately assessing performance in real-world clinical scenarios, as plaque characteristics are not always precisely known.
  • MECT's Advantage: Photon-counting detector-based MECT systems, with their ability to resolve multiple energy bins, demonstrated a clear advantage over DECT and conventional CT, especially when combined with the proposed observer.
  • Reconstruction Matters: The choice of reconstruction filter (FBP-Hanning vs. FBP-Ramp) significantly impacts performance, with Hanning generally yielding better results due to reduced noise.

Key Results

  • MECT Outperforms: MECT systems consistently outperformed DECT and conventional CT systems in both SKE/BKE and SKS/BKE tasks for plaque classification.
  • Significant SNR Improvements: MECT yielded SNR improvements ranging from 46.8% to 67.7% compared to other methods, depending on the reconstruction filter and task.
  • AUC Advantage: The AUC values for MECT were higher across all signal certainty levels and reconstruction methods, indicating superior discrimination capability.
  • Conventional CT Trends: For conventional CT, lower tube voltages (e.g., 80 kVp) generally showed better discrimination performance than higher voltages (120 kVp, 140 kVp), likely due to better material contrast at lower energies.
  • DECT Limitations: DECT systems showed only marginal improvements over conventional CT, suggesting that their limited spectral separation might not be sufficient to significantly enhance plaque characterization beyond what conventional CT can offer.

Limitations and Future Work

  • Simulated Data: The study relied entirely on simulated data. While this allows for controlled experiments and validation, real-world clinical data with its inherent complexities (e.g., imperfect system calibration, patient motion) would be necessary for full validation.
  • Ideal Detector Assumptions: The simulations assumed ideal photon-counting detectors with perfect energy resolution and no pile-up effects. Future work should investigate the impact of non-ideal detector characteristics.
  • Limited Plaque Types: The study focused on binary classification tasks. Extending the framework to multiclass discrimination (e.g., differentiating more than two plaque types simultaneously) is a logical next step.
  • Future Directions: The authors suggest extending the observer to handle volumetric (3D/4D) data, multiclass discrimination, and validating the method with real clinical data, potentially using hybrid images (simulated features in real patient backgrounds). They also propose investigating the impact of varying the number and width of energy bins in MECT.

Summary updated

Abstract

Spectral computed tomography (SCT) generates better image quality than conventional computed tomography (CT). It has overcome several limitations for imaging atherosclerotic plaque. However, the literature evaluating the performance of SCT based on objective image assessment is very limited for the task of discriminating plaques. We developed a numerical-observer method and used it to assess performance on discrimination vulnerable-plaque features and compared the performance among multienergy CT (MECT), dual-energy CT (DECT), and conventional CT methods. Our numerical observer was designed to incorporate all spectral information and comprised two-processing stages. First, each energy-window domain was preprocessed by a set of localized channelized Hotelling observers (CHO). In this step, the spectral image in each energy bin was decorrelated using localized prewhitening and matched filtering with a set of Laguerre-Gaussian channel functions. Second, the series of the intermediate scores computed from all the CHOs were integrated by a Hotelling observer with an additional prewhitening and matched filter. The overall signal-to-noise ratio (SNR) and the area under the receiver operating characteristic curve (AUC) were obtained, yielding an overall discrimination performance metric. The performance of our new observer was evaluated for the particular binary classification task of differentiating between alternative plaque characterizations in carotid arteries. A clinically realistic model of signal variability was also included in our simulation of the discrimination tasks. The inclusion of signal variation is a key to applying the proposed observer method to spectral CT data. Hence, the task-based approaches based on the signal-known-exactly/background-known-exactly (SKE/BKE) framework and the clinical-relevant signal-known-statistically/background-known-exactly (SKS/BKE) framework were applied for analytical computation of figures of merit (FOM). Simulated data of a carotid-atherosclerosis patient were used to validate our methods. We used an extended cardiac-torso anthropomorphic digital phantom and three simulated plaque types (i.e., calcified plaque, fatty-mixed plaque, and iodine-mixed blood). The images were reconstructed using a standard filtered backprojection (FBP) algorithm for all the acquisition methods and were applied to perform two different discrimination tasks of: (1) calcified plaque versus fatty-mixed plaque and (2) calcified plaque versus iodine-mixed blood. MECT outperformed DECT and conventional CT systems for all cases of the SKE/BKE and SKS/BKE tasks (all [Formula: see text]). On average of signal variability, MECT yielded the SNR improvements over other acquisition methods in the range of 46.8% to 65.3% (all [Formula: see text]) for FBP-Ramp images and 53.2% to 67.7% (all [Formula: see text]) for FBP-Hanning images for both identification tasks. This proposed numerical observer combined with our signal variability framework is promising for assessing material characterization obtained through the additional energy-dependent attenuation information of SCT. These methods can be further extended to other clinical tasks such as kidney or urinary stone identification applications.
BibTeX
@article{lorsakul2016numerical,
  author = {Auranuch Lorsakul and Georges El Fakhri and W. Worstell and Jinsong Ouyang and Yothin Rakvongthai and Andrew F. Laine and Quanzheng Li},
  title = {{Numerical observer for atherosclerotic plaque classification in spectral computed tomography}},
  journal = {Journal of Medical Imaging},
  year = {2016},
  doi = {10.1117/1.jmi.3.3.035501},
}
Figure