Evaluating Performance of Microwave Image Reconstruction Algorithms: Extracting Tissue Types with Segmentation Using Machine Learning.

Evaluating the quality of reconstructed images requires consistent approaches to extracting information and applying metrics. Partitioning medical images into tissue types permits the quantitative assessment of regions that contain a specific tissue. The assessment facilitates the evaluation of an imaging algorithm in terms of its ability to reconstruct the properties of various tissue types and identify anomalies. Microwave tomography is an imaging modality that is model-based and reconstructs an approximation of the actual internal spatial distribution of the dielectric properties of a breast over a reconstruction model consisting of discrete elements. The breast tissue types are characterized by their dielectric properties, so the complex permittivity profile that is reconstructed may be used to distinguish different tissue types. This manuscript presents a robust and flexible medical image segmentation technique to partition microwave breast images into tissue types in order to facilitate the evaluation of image quality. The approach combines an unsupervised machine learning method with statistical techniques. The key advantage for using the algorithm over other approaches, such as a threshold-based segmentation method, is that it supports this quantitative analysis without prior assumptions such as knowledge of the expected dielectric property values that characterize each tissue type. Moreover, it can be used for scenarios where there is a scarcity of data available for supervised learning. Microwave images are formed by solving an inverse scattering problem that is severely ill-posed, which has a significant impact on image quality. A number of strategies have been developed to alleviate the ill-posedness of the inverse scattering problem. The degree of success of each strategy varies, leading to reconstructions that have a wide range of image quality. A requirement for the segmentation technique is the ability to partition tissue types over a range of image qualities, which is demonstrated in the first part of the paper. The segmentation of images into regions of interest corresponding to various tissue types leads to the decomposition of the breast interior into disjoint tissue masks. An array of region and distance-based metrics are applied to compare masks extracted from reconstructed images and ground truth models. The quantitative results reveal the accuracy with which the geometric and dielectric properties are reconstructed. The incorporation of the segmentation that results in a framework that effectively furnishes the quantitative assessment of regions that contain a specific tissue is also demonstrated. The algorithm is applied to reconstructed microwave images derived from breasts with various densities and tissue distributions to demonstrate the flexibility of the algorithm and that it is not data-specific. The potential for using the algorithm to assist in diagnosis is exhibited with a tumor tracking example. This example also establishes the usefulness of the approach in evaluating the performance of the reconstruction algorithm in terms of its sensitivity and specificity to malignant tissue and its ability to accurately reconstruct malignant tissue.

Adaptive Monostatic System for Measuring Microwave Reflections from the Breast.

A second-generation monostatic radar system to measure microwave reflections from the human breast is presented and analyzed. The present system can measure the outline of the breast with an accuracy of ±1 mm and precisely place the microwave sensor in an adaptive matter such that microwaves are normally incident on the skin. Microwave reflections are measured between 10 MHz to 12 GHz with sensitivity of 65 to 75 dB below the input power and a total scan time of 30 min for 140 locations. The time domain reflections measured from a volunteer show fidelity above 0.98 for signals in a single scan. Finally, multiple scans of a breast phantoms demonstrate the consistency of the system in terms of recorded reflection, outline measurement, and image reconstruction.

Integrating prior information into microwave tomography Part 1: Impact of detail on image quality.

PURPOSE: The authors investigate the impact that incremental increases in the level of detail of patient-specific prior information have on image quality and the convergence behavior of an inversion algorithm in the context of near-field microwave breast imaging. A methodology is presented that uses image quality measures to characterize the ability of the algorithm to reconstruct both internal structures and lesions embedded in fibroglandular tissue. The approach permits key aspects that impact the quality of reconstruction of these structures to be identified and quantified. This provides insight into opportunities to improve image reconstruction performance. METHODS: Patient-specific information is acquired using radar-based methods that form a regional map of the breast. This map is then incorporated into a microwave tomography algorithm. Previous investigations have demonstrated the effectiveness of this approach to improve image quality when applied to data generated with two-dimensional (2D) numerical models. The present study extends this work by generating prior information that is customized to vary the degree of structural detail to facilitate the investigation of the role of prior information in image formation. Numerical 2D breast models constructed from magnetic resonance (MR) scans, and reconstructions formed with a three-dimensional (3D) numerical breast model are used to assess if trends observed for the 2D results can be extended to 3D scenarios. RESULTS: For the blind reconstruction scenario (i.e., no prior information), the breast surface is not accurately identified and internal structures are not clearly resolved. A substantial improvement in image quality is achieved by incorporating the skin surface map and constraining the imaging domain to the breast. Internal features within the breast appear in the reconstructed image. However, it is challenging to discriminate between adipose and glandular regions and there are inaccuracies in both the structural properties of the glandular region and the dielectric properties reconstructed within this structure. Using a regional map with a skin layer only marginally improves this situation. Increasing the structural detail in the prior information to include internal features leads to reconstructions for which the interface that delineates the fat and gland regions can be inferred. Different features within the glandular region corresponding to tissues with varying relative permittivity values, such as a lesion embedded within glandular structure, emerge in the reconstructed images. CONCLUSION: Including knowledge of the breast surface and skin layer leads to a substantial improvement in image quality compared to the blind case, but the images have limited diagnostic utility for applications such as tumor response tracking. The diagnostic utility of the reconstruction technique is improved considerably when patient-specific structural information is used. This qualitative observation is supported quantitatively with image metrics.

Integrating prior information into microwave tomography part 2: Impact of errors in prior information on microwave tomography image quality.

PURPOSE: The authors have developed a method to combine a patient-specific map of tissue structure and average dielectric properties with microwave tomography. The patient-specific map is acquired with radar-based techniques and serves as prior information for microwave tomography. The impact that the degree of structural detail included in this prior information has on image quality was reported in a previous investigation. The aim of the present study is to extend this previous work by identifying and quantifying the impact that errors in the prior information have on image quality, including the reconstruction of internal structures and lesions embedded in fibroglandular tissue. This study also extends the work of others reported in literature by emulating a clinical setting with a set of experiments that incorporate heterogeneity into both the breast interior and glandular region, as well as prior information related to both fat and glandular structures. METHODS: Patient-specific structural information is acquired using radar-based methods that form a regional map of the breast. Errors are introduced to create a discrepancy in the geometry and electrical properties between the regional map and the model used to generate the data. This permits the impact that errors in the prior information have on image quality to be evaluated. Image quality is quantitatively assessed by measuring the ability of the algorithm to reconstruct both internal structures and lesions embedded in fibroglandular tissue. The study is conducted using both 2D and 3D numerical breast models constructed from MRI scans. RESULTS: The reconstruction results demonstrate robustness of the method relative to errors in the dielectric properties of the background regional map, and to misalignment errors. These errors do not significantly influence the reconstruction accuracy of the underlying structures, or the ability of the algorithm to reconstruct malignant tissue. Although misalignment errors do not significantly impact the quality of the reconstructed fat and glandular structures for the 3D scenarios, the dielectric properties are reconstructed less accurately within the glandular structure for these cases relative to the 2D cases. However, general agreement between the 2D and 3D results was found. CONCLUSION: A key contribution of this paper is the detailed analysis of the impact of prior information errors on the reconstruction accuracy and ability to detect tumors. The results support the utility of acquiring patient-specific information with radar-based techniques and incorporating this information into MWT. The method is robust to errors in the dielectric properties of the background regional map, and to misalignment errors. Completion of this analysis is an important step toward developing the method into a practical diagnostic tool.

Surface Estimation for Microwave Imaging.

Biomedical imaging and sensing applications in many scenarios demand accurate surface estimation from a sparse set of noisy measurements. These measurements may arise from a variety of sensing modalities, including laser or electromagnetic samples of an object's surface. We describe a state-of-the-art microwave imaging prototype that has sensors to acquire both microwave and laser measurements. The approach developed to translate sparse samples of the breast surface into an accurate estimate of the region of interest is detailed. To evaluate the efficacy of the method, laser and electromagnetic samples are acquired by sensors from three realistic breast models with varying sizes and shapes. A set of metrics is developed to assist with the investigation and demonstrate that the algorithm is able to accurately estimate the shape of a realistic breast phantom when only a sparse set of data are available. Moreover, the algorithm is robust to the presence of measurement noise, and is effective when applied to measurement scans of patients acquired with the prototype.

An improved technique to predict the time-of-arrival of a tumor response in radar-based breast imaging.

Radar-based microwave imaging has been proposed as a complementary modality for early-stage breast cancer screening. This paper presents an algorithm that may be used to accurately predict the time-of-arrival (TOA) of a tumor response contained in sample data acquired from a small number of antennas in a realistic scenario. The TOA information may be used by many of the existing radar-based methods to detect and localize a tumor response. The robustness of the algorithm is demonstrated with data generated from realistic numerical breast models.

Tumor response estimation in radar-based microwave breast cancer detection.

Radar-based microwave imaging techniques have been proposed for early stage breast cancer detection. A considerable challenge for the successful implementation of these techniques is the reduction of clutter, or components of the signal originating from objects other than the tumor. In particular, the reduction of clutter from the late-time scattered fields is required in order to detect small (subcentimeter diameter) tumors. In this paper, a method to estimate the tumor response contained in the late-time scattered fields is presented. The method uses a parametric function to model the tumor response. A maximum a posteriori estimation approach is used to evaluate the optimal values for the estimates of the parameters. A pattern classification technique is then used to validate the estimation. The ability of the algorithm to estimate a tumor response is demonstrated by using both experimental and simulated data obtained with a tissue sensing adaptive radar system.

A technique to predict the time-of-arrival of a tumor response corrupted by clutter.

Radar-based microwave imaging has been proposed as a complementary modality for early stage breast cancer screening. A considerable challenge for the successful implementation of this technique is the reduction of clutter, or components of the signal originating from objects other than the tumor. To remedy problems that arise due to the breast shape, size and tissue variations, a novel technique is presented that identifies and localizes the source of a target response. The ability of the algorithm to predict the time-of-arrival of a tumor response in a signal corrupted by clutter is demonstrated.