Transport processes are ubiquitous. They are, for example, at the heart of optical flow approache... more Transport processes are ubiquitous. They are, for example, at the heart of optical flow approaches; or of perfusion imaging, where blood transport is assessed, most commonly by injecting a tracer. An advection-diffusion equation is widely used to describe these transport phenomena. Our goal is estimating the underlying physics of advection-diffusion equations, expressed as velocity and diffusion tensor fields. We propose a learning framework (YETI) building on an auto-encoder structure between 2D and 3D image time-series, which incorporates the advection-diffusion model. To help with identifiability, we develop an advection-diffusion simulator which allows pre-training of our model by supervised learning using the velocity and diffusion tensor fields. Instead of directly learning these velocity and diffusion tensor fields, we introduce representations that assure incompressible flow and symmetric positive semi-definite diffusion fields and demonstrate the additional benefits of thes...
Localization of functional sites across individuals is impor- tant to assess and analyze neurolog... more Localization of functional sites across individuals is impor- tant to assess and analyze neurological disparities in a population. Tradi- tionally this is accomplished by conducting anatomical registration and assuming functional to anatomical relationships. In this paper we in- vestigate the relationship between functional sites and flber tracts. We propose a novel algorithm based on a model-to-image registration scheme to align difiusion tensor flber tracts with fractional anisotropy map and thereby functional sites across individuals. We have evaluated our algo- rithm on ten normal controls, our results show similar accuracy compared to a global image-to-image registration strategy of fractional anisotropy images.
Perfusion imaging (PI) is clinically used to assess strokes and brain tumors. Commonly used PI ap... more Perfusion imaging (PI) is clinically used to assess strokes and brain tumors. Commonly used PI approaches based on magnetic resonance imaging (MRI) or computed tomography (CT) measure the effect of a contrast agent moving through blood vessels and into tissue. Contrast-agent free approaches, for example, based on intravoxel incoherent motion, also exist, but are so far not routinely used clinically. These methods rely on estimating on the arterial input function (AIF) to approximately model tissue perfusion, neglecting spatial dependencies, and reliably estimating the AIF is also non-trivial, leading to difficulties with standardizing perfusion measures. In this work we therefore propose a data-assimilation approach (PIANO) which estimates the velocity and diffusion fields of an advection-diffusion model that best explains the contrast dynamics. PIANO accounts for spatial dependencies and neither requires estimating the AIF nor relies on a particular contrast agent bolus shape. Spec...
In this paper we aim to refine the concept of grand challenges in medical image analysis, based o... more In this paper we aim to refine the concept of grand challenges in medical image analysis, based on statistical principles from quantitative and qualitative experimental research. We identify two types of challenges based on their generalization objective: 1) a deployment challenge and 2) an insight challenge. A deployment challenge's generalization objective is to find algorithms that solve a medical image analysis problem, which thereby requires the use of a quantitative experimental design. An insight challenge's generalization objective is to gain a broad understanding of what class of algorithms might be effective for a class of medical image analysis problems, in which case a qualitative experimental design is sufficient. Both challenge types are valuable, but problems arise when a challenge's design and objective are inconsistent, as is often the case when a challenge does not carefully consider these concepts. Therefore, in this paper, we propose a theoretical fra...
Background: Preclinical ultrasound (US) and contrast-enhanced ultrasound (CEUS) imaging have long... more Background: Preclinical ultrasound (US) and contrast-enhanced ultrasound (CEUS) imaging have long been used in oncology to noninvasively measure tumor volume and vascularity. While the value of preclinical US has been repeatedly demonstrated, these modalities are not without several key limitations that make them unattractive to cancer researchers, including: high user-variability, low throughput, and limited imaging field-of-view (FOV). Herein, we present a novel robotic preclinical US/CEUS system that addresses these limitations and demonstrates its use in evaluating tumors in 3D in a rodent model. Methods: The imaging system was designed to allow seamless whole-body 3D imaging, which requires rodents to be imaged without physical contact between the US transducer and the animal. To achieve this, a custom dual-element transducer was mounted on a robotic carriage, submerged in a hydrocarbon fluid, and the reservoir sealed with an acoustically transmissive top platform. Eight NOD/scid/gamma (NSG) female mice were injected subcutaneously in the flank with 8×109 786-O human clear-cell renal cell carcinoma (ccRCC) cells. Weekly imaging commenced after tumors reached a size of 150 mm3 and continued until tumors reached a maximum size of 1 cm3 (∼4-5 weeks). An additional six nude athymic female mice were injected subcutaneously in the flank with 7 × 105 SVR angiosarcoma cells to perform an inter-operator variability study. Imaging consisted of 3D B-mode (conventional ultrasound) of the whole abdomen ( Results: Wide-field US images reconstructed from 3D volumetric data showed superior FOV over conventional US. Several anatomical landmarks could be identified within each image surrounding the tumor, including the liver, small intestines, bladder, and inguinal lymph nodes. Tumor boundaries were clearly delineated in both B-mode and BVD images, with BVD images showing heterogeneous microvessel density at later timepoints suggesting tumor necrosis. Excellent agreement was measured for both inter-reader and inter-operator experiments, with alpha coefficients of 0.914 (95% CI: 0.824-0.948) and 0.959 (0.911-0.981), respectively. Conclusion: We have demonstrated a novel preclinical US imaging system that can accurately and consistently evaluate tumors in rodent models. The system leverages cost-effective robotic technology, and a new scanning paradigm that allows for easy and reproducible data acquisition to enable wide-field, 3D, multi-parametric ultrasound imaging. Note: This abstract was not presented at the meeting. Citation Format: Tomasz Czernuszewicz, Virginie Papadopoulou, Juan D. Rojas, Rajalekha Rajamahendiran, Jonathan Perdomo, James Butler, Max Harlacher, Graeme O9Connell, Dzenan Zukic, Paul A. Dayton, Stephen Aylward, Ryan C. Gessner. A preclinical ultrasound platform for widefield 3D imaging of rodent tumors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1955.
IEEE transactions on bio-medical engineering, Jan 27, 2018
Functional and molecular changes often precede gross anatomical changes, so early assessment of a... more Functional and molecular changes often precede gross anatomical changes, so early assessment of a tumor's functional and molecular response to therapy can help reduce a patient's exposure to the side effects of ineffective chemotherapeutics or other treatment strategies. Our intent was to test the hypothesis that an ultrasound microvascular imaging approach might provide indications of response to therapy prior to assessment of tumor size. Mice bearing clear-cell renal cell carcinoma xenograft tumors were treated with antiangiogenic and Notch inhibition therapies. An ultrasound measurement of microvascular density was used to serially track the tumor response to therapy. Data indicated that ultrasound-derived microvascular density can indicate response to therapy a week prior to changes in tumor volume and is strongly correlated with physiological characteristics of the tumors as measured by histology (p=0.75). Furthermore, data demonstrated that ultrasound measurements of v...
Brain extraction from 3D medical images is a common pre-processing step. A variety of approaches ... more Brain extraction from 3D medical images is a common pre-processing step. A variety of approaches exist, but they are frequently only designed to perform brain extraction from images without strong pathologies. Extracting the brain from images exhibiting strong pathologies, for example, the presence of a brain tumor or of a traumatic brain injury (TBI), is challenging. In such cases, tissue appearance may substantially deviate from normal tissue appearance and hence violates algorithmic assumptions for standard approaches to brain extraction; consequently, the brain may not be correctly extracted. This paper proposes a brain extraction approach which can explicitly account for pathologies by jointly modeling normal tissue appearance and pathologies. Specifically, our model uses a three-part image decomposition: (1) normal tissue appearance is captured by principal component analysis (PCA), (2) pathologies are captured via a total variation term, and (3) the skull and surrounding tiss...
Simulation and synthesis in medical imaging : first International Workshop, SASHIMI 2016, held in conjunction with MICCAI 2016, Athens, Greece, October 21, 2016, Proceedings. SASHIMI (Workshop) (1st : 2016 : Athens, Greece), 2016
This paper proposes an approach to improve atlas-to-image registration accuracy with large pathol... more This paper proposes an approach to improve atlas-to-image registration accuracy with large pathologies. Instead of directly registering an atlas to a pathological image, the method learns a mapping from the pathological image to a quasi-normal image, for which more accurate registration is possible. Specifically, the method uses a deep variational convolutional encoder-decoder network to learn the mapping. Furthermore, the method estimates local mapping uncertainty through network inference statistics and uses those estimates to down-weight the image registration similarity measure in areas of high uncertainty. The performance of the method is quantified using synthetic brain tumor images and images from the brain tumor segmentation challenge (BRATS 2015).
Proceedings. IEEE International Symposium on Biomedical Imaging, 2017
Registration involving one or more images containing pathologies is challenging, as standard imag... more Registration involving one or more images containing pathologies is challenging, as standard image similarity measures and spatial transforms cannot account for common changes due to pathologies. Low-rank/Sparse (LRS) decomposition removes pathologies prior to registration; however, LRS is memory-demanding and slow, which limits its use on larger data sets. Additionally, LRS blurs normal tissue regions, which may degrade registration performance. This paper proposes an efficient alternative to LRS: (1) normal tissue appearance is captured by principal component analysis (PCA) and (2) blurring is avoided by an integrated model for pathology removal and image reconstruction. Results on synthetic and BRATS 2015 data demonstrate its utility.
Proceedings. IEEE International Symposium on Biomedical Imaging, 2018
We aim to diagnose scoliosis using a self contained ultrasound device that does not require signi... more We aim to diagnose scoliosis using a self contained ultrasound device that does not require significant training to operate. The device knows its angle relative to vertical using an embedded inertial measurement unit, and it estimates its angle relative to a vertebrae using a neural network analysis of its ultrasound images. The composition of those angles defines the angle of a vertebrae from vertical. The maximum difference between vertebrae angles collected from a scan of a spine yields the Cobb angle measure that is used to quantify scoliosis severity.
We introduce an automated and accurate system for reg- istering pre-operative 3D MR and CT images... more We introduce an automated and accurate system for reg- istering pre-operative 3D MR and CT images with intra- operative 3D ultrasound images based on the vessels visible in both. The clinical goal is to guide the radio-frequency ab- lation (RFA) of liver lesions using percutaneous ultrasound even when the lesions are not directly visible using ultra- sound. The lesions locations
The National Alliance for Medical Image Computing (NA-MIC) was launched in 2004 with the goal of ... more The National Alliance for Medical Image Computing (NA-MIC) was launched in 2004 with the goal of investigating and developing an open source software infrastructure for the extraction of information and knowledge from medical images using computational methods. Several leading research and engineering groups participated in this effort that was funded by the US National Institutes of Health through a variety of infrastructure grants. This effort transformed 3D Slicer from an internal, Boston-based, academic research software application into a professionally maintained, robust, open source platform with an international leadership and developer and user communities. Critical improvements to the widely used underlying open source libraries and tools-VTK, ITK, CMake, CDash, DCMTK-were an additional consequence of this effort. This project has contributed to close to a thousand peer-reviewed publications and a growing portfolio of US and international funded efforts expanding the use of these tools in new medical computing applications every year. In this editorial, we discuss what we believe are gaps in the way medical image computing is pursued today; how a well-executed research platform can enable discovery, innovation and reproducible science ("Open Science"); and how our quest to build such a software platform has evolved into a productive and rewarding social engineering exercise in building an open-access community with a shared vision.
Proceedings of Spie the International Society For Optical Engineering, 1999
[Proceedings of SPIE 3658, 382 (1999)]. Bradley M. Hemminger, Elodia Cole, Shuquan Zong, Martin J... more [Proceedings of SPIE 3658, 382 (1999)]. Bradley M. Hemminger, Elodia Cole, Shuquan Zong, Martin J. Yaffe, Stephen Aylward, Andrew DA Maidment, Mark B. Williams, Loren T. Niklason, Richard E. Johnston, Etta D. Pisano. Abstract. ...
ABSTRACT We present a novel method forrepresenting "extruded" distributions. An... more ABSTRACT We present a novel method forrepresenting "extruded" distributions. An extruded distribution is an M-dimensional manifoldin the parameter space of the component distribution. Representations of that manifoldare "continuous mixture models". We present a method forforming one-dimensional continuous Gaussian mixture models of sampled extruded Gaussian distributions viaridgk of gOBOLEd"qkOLUd Using Monte Carlo simulations and ROC analysis, we explore the utility of a variety ofbinning techniques andgd,qkEDd"qqU, functions. We demonstrate that extruded Gaussian distributions are more accurately andconsistently representedby continuous Gaussian mixture models than by #nite Gaussian mixture models formedvia maximum likelihoodexpectation maximization. ? 2002 PatternRecogdkkEU Society. Published by Elsevier Science Ltd. Allrig,, reserved.
Transport processes are ubiquitous. They are, for example, at the heart of optical flow approache... more Transport processes are ubiquitous. They are, for example, at the heart of optical flow approaches; or of perfusion imaging, where blood transport is assessed, most commonly by injecting a tracer. An advection-diffusion equation is widely used to describe these transport phenomena. Our goal is estimating the underlying physics of advection-diffusion equations, expressed as velocity and diffusion tensor fields. We propose a learning framework (YETI) building on an auto-encoder structure between 2D and 3D image time-series, which incorporates the advection-diffusion model. To help with identifiability, we develop an advection-diffusion simulator which allows pre-training of our model by supervised learning using the velocity and diffusion tensor fields. Instead of directly learning these velocity and diffusion tensor fields, we introduce representations that assure incompressible flow and symmetric positive semi-definite diffusion fields and demonstrate the additional benefits of thes...
Localization of functional sites across individuals is impor- tant to assess and analyze neurolog... more Localization of functional sites across individuals is impor- tant to assess and analyze neurological disparities in a population. Tradi- tionally this is accomplished by conducting anatomical registration and assuming functional to anatomical relationships. In this paper we in- vestigate the relationship between functional sites and flber tracts. We propose a novel algorithm based on a model-to-image registration scheme to align difiusion tensor flber tracts with fractional anisotropy map and thereby functional sites across individuals. We have evaluated our algo- rithm on ten normal controls, our results show similar accuracy compared to a global image-to-image registration strategy of fractional anisotropy images.
Perfusion imaging (PI) is clinically used to assess strokes and brain tumors. Commonly used PI ap... more Perfusion imaging (PI) is clinically used to assess strokes and brain tumors. Commonly used PI approaches based on magnetic resonance imaging (MRI) or computed tomography (CT) measure the effect of a contrast agent moving through blood vessels and into tissue. Contrast-agent free approaches, for example, based on intravoxel incoherent motion, also exist, but are so far not routinely used clinically. These methods rely on estimating on the arterial input function (AIF) to approximately model tissue perfusion, neglecting spatial dependencies, and reliably estimating the AIF is also non-trivial, leading to difficulties with standardizing perfusion measures. In this work we therefore propose a data-assimilation approach (PIANO) which estimates the velocity and diffusion fields of an advection-diffusion model that best explains the contrast dynamics. PIANO accounts for spatial dependencies and neither requires estimating the AIF nor relies on a particular contrast agent bolus shape. Spec...
In this paper we aim to refine the concept of grand challenges in medical image analysis, based o... more In this paper we aim to refine the concept of grand challenges in medical image analysis, based on statistical principles from quantitative and qualitative experimental research. We identify two types of challenges based on their generalization objective: 1) a deployment challenge and 2) an insight challenge. A deployment challenge's generalization objective is to find algorithms that solve a medical image analysis problem, which thereby requires the use of a quantitative experimental design. An insight challenge's generalization objective is to gain a broad understanding of what class of algorithms might be effective for a class of medical image analysis problems, in which case a qualitative experimental design is sufficient. Both challenge types are valuable, but problems arise when a challenge's design and objective are inconsistent, as is often the case when a challenge does not carefully consider these concepts. Therefore, in this paper, we propose a theoretical fra...
Background: Preclinical ultrasound (US) and contrast-enhanced ultrasound (CEUS) imaging have long... more Background: Preclinical ultrasound (US) and contrast-enhanced ultrasound (CEUS) imaging have long been used in oncology to noninvasively measure tumor volume and vascularity. While the value of preclinical US has been repeatedly demonstrated, these modalities are not without several key limitations that make them unattractive to cancer researchers, including: high user-variability, low throughput, and limited imaging field-of-view (FOV). Herein, we present a novel robotic preclinical US/CEUS system that addresses these limitations and demonstrates its use in evaluating tumors in 3D in a rodent model. Methods: The imaging system was designed to allow seamless whole-body 3D imaging, which requires rodents to be imaged without physical contact between the US transducer and the animal. To achieve this, a custom dual-element transducer was mounted on a robotic carriage, submerged in a hydrocarbon fluid, and the reservoir sealed with an acoustically transmissive top platform. Eight NOD/scid/gamma (NSG) female mice were injected subcutaneously in the flank with 8×109 786-O human clear-cell renal cell carcinoma (ccRCC) cells. Weekly imaging commenced after tumors reached a size of 150 mm3 and continued until tumors reached a maximum size of 1 cm3 (∼4-5 weeks). An additional six nude athymic female mice were injected subcutaneously in the flank with 7 × 105 SVR angiosarcoma cells to perform an inter-operator variability study. Imaging consisted of 3D B-mode (conventional ultrasound) of the whole abdomen ( Results: Wide-field US images reconstructed from 3D volumetric data showed superior FOV over conventional US. Several anatomical landmarks could be identified within each image surrounding the tumor, including the liver, small intestines, bladder, and inguinal lymph nodes. Tumor boundaries were clearly delineated in both B-mode and BVD images, with BVD images showing heterogeneous microvessel density at later timepoints suggesting tumor necrosis. Excellent agreement was measured for both inter-reader and inter-operator experiments, with alpha coefficients of 0.914 (95% CI: 0.824-0.948) and 0.959 (0.911-0.981), respectively. Conclusion: We have demonstrated a novel preclinical US imaging system that can accurately and consistently evaluate tumors in rodent models. The system leverages cost-effective robotic technology, and a new scanning paradigm that allows for easy and reproducible data acquisition to enable wide-field, 3D, multi-parametric ultrasound imaging. Note: This abstract was not presented at the meeting. Citation Format: Tomasz Czernuszewicz, Virginie Papadopoulou, Juan D. Rojas, Rajalekha Rajamahendiran, Jonathan Perdomo, James Butler, Max Harlacher, Graeme O9Connell, Dzenan Zukic, Paul A. Dayton, Stephen Aylward, Ryan C. Gessner. A preclinical ultrasound platform for widefield 3D imaging of rodent tumors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1955.
IEEE transactions on bio-medical engineering, Jan 27, 2018
Functional and molecular changes often precede gross anatomical changes, so early assessment of a... more Functional and molecular changes often precede gross anatomical changes, so early assessment of a tumor's functional and molecular response to therapy can help reduce a patient's exposure to the side effects of ineffective chemotherapeutics or other treatment strategies. Our intent was to test the hypothesis that an ultrasound microvascular imaging approach might provide indications of response to therapy prior to assessment of tumor size. Mice bearing clear-cell renal cell carcinoma xenograft tumors were treated with antiangiogenic and Notch inhibition therapies. An ultrasound measurement of microvascular density was used to serially track the tumor response to therapy. Data indicated that ultrasound-derived microvascular density can indicate response to therapy a week prior to changes in tumor volume and is strongly correlated with physiological characteristics of the tumors as measured by histology (p=0.75). Furthermore, data demonstrated that ultrasound measurements of v...
Brain extraction from 3D medical images is a common pre-processing step. A variety of approaches ... more Brain extraction from 3D medical images is a common pre-processing step. A variety of approaches exist, but they are frequently only designed to perform brain extraction from images without strong pathologies. Extracting the brain from images exhibiting strong pathologies, for example, the presence of a brain tumor or of a traumatic brain injury (TBI), is challenging. In such cases, tissue appearance may substantially deviate from normal tissue appearance and hence violates algorithmic assumptions for standard approaches to brain extraction; consequently, the brain may not be correctly extracted. This paper proposes a brain extraction approach which can explicitly account for pathologies by jointly modeling normal tissue appearance and pathologies. Specifically, our model uses a three-part image decomposition: (1) normal tissue appearance is captured by principal component analysis (PCA), (2) pathologies are captured via a total variation term, and (3) the skull and surrounding tiss...
Simulation and synthesis in medical imaging : first International Workshop, SASHIMI 2016, held in conjunction with MICCAI 2016, Athens, Greece, October 21, 2016, Proceedings. SASHIMI (Workshop) (1st : 2016 : Athens, Greece), 2016
This paper proposes an approach to improve atlas-to-image registration accuracy with large pathol... more This paper proposes an approach to improve atlas-to-image registration accuracy with large pathologies. Instead of directly registering an atlas to a pathological image, the method learns a mapping from the pathological image to a quasi-normal image, for which more accurate registration is possible. Specifically, the method uses a deep variational convolutional encoder-decoder network to learn the mapping. Furthermore, the method estimates local mapping uncertainty through network inference statistics and uses those estimates to down-weight the image registration similarity measure in areas of high uncertainty. The performance of the method is quantified using synthetic brain tumor images and images from the brain tumor segmentation challenge (BRATS 2015).
Proceedings. IEEE International Symposium on Biomedical Imaging, 2017
Registration involving one or more images containing pathologies is challenging, as standard imag... more Registration involving one or more images containing pathologies is challenging, as standard image similarity measures and spatial transforms cannot account for common changes due to pathologies. Low-rank/Sparse (LRS) decomposition removes pathologies prior to registration; however, LRS is memory-demanding and slow, which limits its use on larger data sets. Additionally, LRS blurs normal tissue regions, which may degrade registration performance. This paper proposes an efficient alternative to LRS: (1) normal tissue appearance is captured by principal component analysis (PCA) and (2) blurring is avoided by an integrated model for pathology removal and image reconstruction. Results on synthetic and BRATS 2015 data demonstrate its utility.
Proceedings. IEEE International Symposium on Biomedical Imaging, 2018
We aim to diagnose scoliosis using a self contained ultrasound device that does not require signi... more We aim to diagnose scoliosis using a self contained ultrasound device that does not require significant training to operate. The device knows its angle relative to vertical using an embedded inertial measurement unit, and it estimates its angle relative to a vertebrae using a neural network analysis of its ultrasound images. The composition of those angles defines the angle of a vertebrae from vertical. The maximum difference between vertebrae angles collected from a scan of a spine yields the Cobb angle measure that is used to quantify scoliosis severity.
We introduce an automated and accurate system for reg- istering pre-operative 3D MR and CT images... more We introduce an automated and accurate system for reg- istering pre-operative 3D MR and CT images with intra- operative 3D ultrasound images based on the vessels visible in both. The clinical goal is to guide the radio-frequency ab- lation (RFA) of liver lesions using percutaneous ultrasound even when the lesions are not directly visible using ultra- sound. The lesions locations
The National Alliance for Medical Image Computing (NA-MIC) was launched in 2004 with the goal of ... more The National Alliance for Medical Image Computing (NA-MIC) was launched in 2004 with the goal of investigating and developing an open source software infrastructure for the extraction of information and knowledge from medical images using computational methods. Several leading research and engineering groups participated in this effort that was funded by the US National Institutes of Health through a variety of infrastructure grants. This effort transformed 3D Slicer from an internal, Boston-based, academic research software application into a professionally maintained, robust, open source platform with an international leadership and developer and user communities. Critical improvements to the widely used underlying open source libraries and tools-VTK, ITK, CMake, CDash, DCMTK-were an additional consequence of this effort. This project has contributed to close to a thousand peer-reviewed publications and a growing portfolio of US and international funded efforts expanding the use of these tools in new medical computing applications every year. In this editorial, we discuss what we believe are gaps in the way medical image computing is pursued today; how a well-executed research platform can enable discovery, innovation and reproducible science ("Open Science"); and how our quest to build such a software platform has evolved into a productive and rewarding social engineering exercise in building an open-access community with a shared vision.
Proceedings of Spie the International Society For Optical Engineering, 1999
[Proceedings of SPIE 3658, 382 (1999)]. Bradley M. Hemminger, Elodia Cole, Shuquan Zong, Martin J... more [Proceedings of SPIE 3658, 382 (1999)]. Bradley M. Hemminger, Elodia Cole, Shuquan Zong, Martin J. Yaffe, Stephen Aylward, Andrew DA Maidment, Mark B. Williams, Loren T. Niklason, Richard E. Johnston, Etta D. Pisano. Abstract. ...
ABSTRACT We present a novel method forrepresenting "extruded" distributions. An... more ABSTRACT We present a novel method forrepresenting "extruded" distributions. An extruded distribution is an M-dimensional manifoldin the parameter space of the component distribution. Representations of that manifoldare "continuous mixture models". We present a method forforming one-dimensional continuous Gaussian mixture models of sampled extruded Gaussian distributions viaridgk of gOBOLEd"qkOLUd Using Monte Carlo simulations and ROC analysis, we explore the utility of a variety ofbinning techniques andgd,qkEDd"qqU, functions. We demonstrate that extruded Gaussian distributions are more accurately andconsistently representedby continuous Gaussian mixture models than by #nite Gaussian mixture models formedvia maximum likelihoodexpectation maximization. ? 2002 PatternRecogdkkEU Society. Published by Elsevier Science Ltd. Allrig,, reserved.
Uploads
Papers by Stephen Aylward