Home - djkurran/Automated-framework-for-evaluating-microwave-and-multi-modality-breast-images GitHub Wiki
Automated-workflow-for-evaluating-microwave-and-multi-modality-breast images
Abstract
The emergence and subsequent expansion of the field of medical microwave imaging has resulted in numerous approaches to image reconstruction. This includes microwave tomography, radar imaging, and more recently, multi-modality approaches. However, there is an absence of a standardized and widely accepted process that is proficient at extracting information from these images and employing this knowledge to conduct a thorough quantitative evaluation of images and regions within images. This shortcoming may interfere with a researcher’s ability to make reliable and consistent inferences from experiments and to interpret results. Consequently, comparing the results of different research groups is difficult. This is becoming increasingly relevant due to the development of standardized test phantoms and the increase in clinical studies. To remedy this deficiency, an automated workflow has been developed with the objective to standardize the processing and analysis of images acquired from a range of modalities. Images are first segmented into regions dominated by a tissue type. Quantitative information extracted from these regions is used for analysis and by visualization tools for the qualitative interpretation of images. The effectiveness of the workflow is demonstrated with multiple examples that focus on quantifying changes to images due to enhancements of the reconstruction algorithm or perturbations of a parameter used by the reconstruction operator.
Overview and take-home messages
Figure 1. Overview of automated workflow for evaluating microwave and multi-modality breast images.
Overview
The objective is to furnish a workflow that standardizes the processing and analysis of images to allow researchers to make inferences from experiments and to report consistent and reliable results that can be compared across research groups. The workflow is applied to pairs of images: a test, and a reference image. The test image is a reconstruction, but the reference image may be another version of the reconstruction or a ground truth model.
Images are segmented into regions dominated by a tissue type. Quantitative information is then extracted from these regions. This is accomplished by comparing the test image to the reference. Discrepancies are measured by comparing regions based on shape, size, geometric features, and location within an image. This is carried out by the Tissue interface analysis, and geometric analysis components of the workflow. Moreover, properties within segmented regions of the test image are compared to the corresponding reference region to evaluate a change in response or the ability of an algorithm to reconstruct a tissue type. This is prescribed by the Analysis of properties stage of the workflow. The properties contained within the segmented regions may be dielectric properties associated with a tissue type (microwave tomography), backscattered energy related to a dominant scatterer (microwave radar), or scattering that may arise due to the skin surface or the presence of dense tissue (qualitative ultrasound).
Take-home messages
- An automated image analysis workflow is presented that extracts quantitative information from regions, objects, and responses within images. The information is used for quantitative analysis and by visualization tools for the qualitative interpretation of images.
- Images are acquired from a range of modalities including tomography, radar, ultrasound, and multi-modality approaches. The workflow is applied to image pairs: a test and a reference. Images are partitioned into regions based on tissue type. Differences between the test-reference image pair are measured by comparing regions attributes including shape, size, geometric features, and location.
- Properties within segmented regions of the test image are compared to the corresponding reference region to measure a change in response or the ability of an algorithm to reconstruct a tissue type.
- Quantitative information extracted from results about an experiment assists researchers to make inferences related to a variable change, a data acquisition modification, or a reconstruction algorithm enhancement, and to facilitate the comparison of results obtained by different research groups.
Introduction
An automated workflow is presented whereby images acquired from a range of modalities are processed to extract quantitative information about an experiment conducted by a researcher. Images are first segmented into regions dominated by a tissue type. Quantitative information is then evaluated from these individual regions within the acquired images. Figure 1 shows an overview of the workflow. The segmentation process is summarized in section 1. A more detailed description is provided with each presented example, as the specific methodology implemented is dependent on the modality used to acquire the data and the reconstruction algorithm used to form the images that are being analyzed. The metrics that are applied to the segmented regions to extract quantitative image information are also described in section 1.
A number of examples are presented to demonstrate the application of the workflow to microwave tomography, microwave radar and ultrasound images. For the examples presented in sections 2-4, researchers investigate the use of dual-modality breast imaging methods combining ultrasound and microwaves. A series of numerical experiments are conducted to investigate the utility of incorporating internal breast structural information acquired from ultrasound imagery into the reconstruction methodology that is applied to microwave data. Accordingly, the workflow is used to measure and quantify changes to a reconstruction that arise by using the dual-modality approach. Both microwave radar and tomography images are segmented and analyzed.
The reconstruction methodology is fixed in the 3D examples presented in sections 5-6. The workflow is adapted to operate on volumes and 3D masks, representing responses (termed response objects). The workflow also accommodates the ability to compare and quantify differences between two reconstructions. Here, the test and reference images are set to two different reconstructions. Application of the workflow to this test-reference image pair quantifies changes within the test image relative to the reference. Experiments are carried out by perturbing a parameter implemented by the reconstruction operator that is applied to microwave reflection data. The workflow is then used to evaluate the impact that a parameter change has on an independent variable such as a shift or smearing of a response within the reconstructed image.
In sections 7-8, the workflow is applied to images reconstructed from numerical ultrasound reflection data. The aim of the reconstruction algorithm is to extract knowledge related to the breast’s internal structure from ultrasound reflection data, and to incorporate this prior structural information into the reconstruction algorithm implemented for microwave imaging. For this scenario, the workflow is used to evaluate the effectiveness of the reconstruction algorithm to identify the breast surface and the delineation of the dense tissue regions.
Finally, conclusions are presented in section 9.
Table of Contents
2. Application to microwave tomography: Heterogeneously dense breast example
3. Application to microwave radar images: Fatty breast example
4. Application to microwave radar images: Dense breast example
5. Application to 3D microwave radar images: Measuring changes within a reconstruction (Case 1)
6. Application to 3D microwave radar images: Measuring changes within a reconstruction (Case 2)
7. Application to ultrasound images: Fatty breast example