Premkumar Elangovan

Simulation and assessment of realistic breast lesions using fractal growth models

A new method of generating realistic three dimensional simulated breast lesions known as diffusion limited aggregation (DLA) is presented, and compared with the random walk (RW) method. Both methods of lesion simulation utilize a physics-based method for inserting these simulated lesions into 2D clinical mammogram images that takes into account the polychromatic x-ray spectrum, local glandularity and scatter. DLA and RW masses were assessed for realism via a receiver operating characteristic (ROC) study with nine observers. The study comprised 150 images of which 50 were real pathology proven mammograms, 50 were normal mammograms with RW inserted masses and 50 were normal mammograms with DLA inserted masses. The average area under the ROC curve for the DLA method was 0.55 (95% confidence interval 0.51-0.59) compared to 0.60 (95% confidence interval 0.56-0.63) for the RW method. The observer study results suggest that the DLA method produced more realistic masses with more variability in shape compared to the RW method. DLA generated lesions can overcome the lack of complexity in structure and shape in many current methods of mass simulation.

Assessment of Microsoft Kinect technology (Kinect for Xbox and Kinect for windows) for patient monitoring during external beam radiotherapy

In external beam radiotherapy, patient misalignment during set-up and motion during treatment may result in lost dose to target tissue and increased dose to normal tissues, reducing therapeutic benefit. The most common method for initial patient setup uses room mounted lasers and surface marks on the skin. We propose to use the Microsoft Kinect which can capture a complete patient skin surface representing a multiplicity of 3D points in a fast reproducible, marker-less manner. Our first experiments quantitatively assess the technical performance of Kinect technology using a planar test object and a precision motion platform to compare the performance of Kinect for Xbox and Kinect for Windows. Further experiments were undertaken to investigate the likely performance of using the Kinect during treatment to detect respiratory motion, both in supine and prone positions. The Windows version of the Kinect produces superior performance of less than 2mm mean error at 80-100 cm distance.

A fast scatter field estimator for digital breast tomosynthesis

Digital breast tomosynthesis (DBT) is a promising alternative approach to overcome the limitations of tissue superposition found in full-field 2D digital mammography. However, due to the absence of anti-scatter grids in DBT, accurate scatter estimation for each projection is necessary for modelling the image reconstruction stage. In this work we identify the limitations associated with scatter estimation using spatial invariant scatter kernels, in particular at the edge region where such methods result in scatter overestimation. Such approaches show an overestimation of scatter-to-primary ratio of over 50% at the edges when compared with results from direct Monte Carlo simulation. This problem was found to increase with projection angle. Simulation work presented here shows that this overestimation in scatter is largely due to air gap between the lower curved breast edge and the detector. We propose a new fast, accurate scatter field estimator for use in DBT which not only considers the breast thickness and primary incidence angle, but also accounts for scatter exiting the breast edge region and traversing an air gap prior to absorption in the detector. The new proposed scatter estimator represents an alternative approach to this problem which reduces discrepancies at the edge of a breast phantom. Moreover, the time required for generating scatter has dropped from approximately 12 hours using Monte Carlo simulations for 1010 photons to just a few minutes per projection. The insertion of scatter from the compression paddle to aforementioned methodologies is also discussed.

High throughput screening for mammography using a human-computer interface with rapid serial visual presentation (RSVP)

The steady rise of the breast cancer screening population, coupled with data expansion produced by new digital screening technologies (tomosynthesis/CT) motivates the development of new, more efficient image screening processes. Rapid Serial Visual Presentation (RSVP) is a new fast-content recognition approach which uses electroencephalography to record brain activity elicited by fast bursts of image data. These brain responses are then subjected to machine classification methods to reveal the expert’s ‘reflex’ response to classify images according to their presence or absence of particular targets. The benefit of this method is that images can be presented at high temporal rates (~10 per second), faster than that required for fully conscious detection, facilitating a high throughput of image (screening) material. In the present paper we present the first application of RSVP to medical image data, and demonstrate how cortically coupled computer vision can be successfully applied to breast cancer screening. Whilst prior RSVP work has utilised multichannel approaches, we also present the first RSVP results demonstrating discriminatory response on a single electrode with a ROC area under the curve of 0.62- 0.86 using a simple Fisher discriminator for classification. This increases to 0.75 – 0.94 when multiple electrodes are used in combination.

Performance comparison of breast imaging modalities using a 4AFC human observer study

This work compares the visibility of spheres and simulated masses in 2D-mammography and tomosynthesis systems using human observer studies. Performing comparison studies between breast imaging systems poses a number of practical challenges within a clinical environment. We therefore adopted a simulation approach which included synthetic breast blocks, a validated lesion simulation model and a set of validated image modelling tools as a viable alternative to clinical trials. A series of 4-alternative forced choice (4AFC) human observer experiments has been conducted for signal detection tasks using masses and spheres as targets. Five physicists participated in the study viewing images with a 5mm target at a range of contrast levels and 60 trials per experimental condition. The results showed that tomosynthesis has a lower threshold contrast than 2D-mammography for masses and spheres, and that detection studies using spheres may produce overly-optimistic threshold contrast values.

Realistic simulation of breast mass appearance using random walk

The aim of the present work was to develop a method for simulating breast lesions in digital mammographic images. Based on the visual appearance of real masses, three dimensional masses were created using a 3D random walk method where the choice of parameters (number of walks and number of steps) enables one to control the appearance of the simulated structure. This work is the first occasion that the random walk results have been combined with a model of digital mammographic imaging systems. This model takes into account appropriate physical image acquisition processes representing a particular digital X-ray mammography system. The X-ray spectrum, local glandularity above the insertion site and scatter were all taken account during the insertion procedure. A preliminary observer study was used to validate the realism of the masses. Seven expert readers each viewed 60 full field mammograms and rated the realism of the masses they contained. Half of the images contained real, histologically-confirmed masses, and half contained simulated lesions. The ROC analysis of the study (average AUC of 0.58±0.06) suggests that, on the average, there is evidence that the radiologists could distinguish, somewhat, between real and simulated masses.

Simulation of 3D DLA masses in digital breast tomosynthesis

Digital breast tomosynthesis (DBT) is suggested to have superior performance compared to 2D mammography in terms of cancer visibility, especially in the case of dense breasts. However, the overall performance of tomosynthesis for screening applications, and the manner in which tomosynthesis should be optimally used for screening remains unclear. This motivates the development of software tools that can insert user-defined synthetic pathology of realistic appearance into clinical tomosynthesis images for subsequent use in virtual clinical trials. We present a method for inserting lesions grown using Diffusion Limited Aggregation, previously validated in 2D mammograms, into clinical DBT images. A preliminary pilot study was used to validate the realism of the masses, wherein three readers each viewed 19 cases and rated the realism of the inserted masses. Each case included a simulated mass inserted in the tomosynthesis projections and the counterpart digital 2D mammogram. These results show that masses can be successfully embedded in the tomosynthesis projections and can produce visually authentic DBT images containing synthetic pathology. These results will be used to further optimize the appearance of these masses in DBT for an upcoming validation.

A modelling framework for evaluation of 2d-mammography and breast tomosynthesis systems

Planar 2D X-ray mammography is the most common screening technique used for breast cancer detection. Digital breast tomosynthesis (DBT) is a new and emerging technology that overcomes some of the limitations of conventional planar imaging. However, it is important to understand the impact of these two modalities on cancer detection rates and patient recall. Since it is difficult to adequately evaluate different modalities clinically, a collection of modeling tools is introduced in this paper that can be used to emulate the image acquisition process for both modalities. In this paper, we discuss image simulation chains that can be used for the evaluation of 2D-mammography and DBT systems in terms of both technical factors and observer studies.

Virtual clinical trials using inserted pathology in clinical images: investigation of assumptions for local glandularity and noise

Virtual clinical trials have been proposed as a viable alternative to clinical trials for testing and comparing the performance of breast imaging systems. One of the main simulation methodologies used in virtual trials employs clinical images of patients in which simulated models of cancer are inserted using a physics-based template multiplication technique. The purpose of this work is to investigate two assumptions commonly considered in this simulation approach: Firstly, given the absence of useful depth information in a clinical situation, an average measure of the local breast glandularity is commonly used as an estimate of the breast composition at the insertion site; secondly, it is also assumed that any change in the relative noise in the image at the insertion site, after insertion of a mass, is negligible. In order to test the validity of these assumptions, spheres representing idealised masses and anthropomorphic computational breast phantoms with perfect prior knowledge of local tissue composition and distribution were used. Results from several region of interest (ROI) insertions demonstrated a lack of variation obtained in contrast with insertion depth using the template multiplication insertion method as compared to the true depth-wise variation contrast values obtained from voxel replacement in a heterogeneous phantom. It was also found that the amount of noise is underestimated by insertion of spherical masses using template multiplication method by 8% - 29% compared to voxel replacement for the test conditions. This resulted in up to 12% variation in contrast-to-noise-ratio (CNR) values between template multiplication and voxel replacement methods.

The effect of system geometry and dose on the threshold detectable calcification diameter in 2D-mammography and digital breast tomosynthesis.

Digital breast tomosynthesis (DBT) is under consideration to replace or to be used in combination with 2D-mammography in breast screening. The aim of this study was the comparison of the detection of microcalcification clusters by human observers in simulated breast images using 2D-mammography, narrow angle (15°/15 projections) and wide angle (50°/25 projections) DBT. The effects of the cluster height in the breast and the dose to the breast on calcification detection were also tested. Simulated images of 6 cm thick compressed breasts were produced with and without microcalcification clusters inserted, using a set of image modelling tools for 2D-mammography and DBT. Image processing and reconstruction were performed using commercial software. A series of 4-alternative forced choice (4AFC) experiments was conducted for signal detection with the microcalcification clusters as targets. Threshold detectable calcification diameter was found for each imaging modality with standard dose: 2D-mammography: 2D-mammography (165  ±  9 µm), narrow angle DBT (211  ±  11 µm) and wide angle DBT (257  ±  14 µm). Statistically significant differences were found when using different doses, but different geometries had a greater effect. No differences were found between the threshold detectable calcification diameters at different heights in the breast. Calcification clusters may have a lower detectability using DBT than 2D imaging.