CMU  


Deep Learning for Microwave Image Reconstruction
 

The traditional image reconstruction methods for microwave imaging such as the constrast source inversion (CSI) are prone to producing artifacts. This project presents a proof-of-concept for a CNN architecture employed for breast image reconstruction from microwave imaging. The network consists of an encoder-decoder architecture involving successive convolutional and downsampling layers, followed by successve deconvolutional and upsampling layers. The input to the network is a rough reconstructions from CSI output, which is mapped to the real and imaginary parts of the complex valued permittivity.,

V. Khoshdel, A. Ashraf, J. LoVetri, "Enhancement of Multimodal Microwave-Ultrasound Breast Imaging using a Deep-Learning Technique", Sensors, 2019 [Link]


Spatiotemporal Heterogeneity Encoding Features for MRI Perfusion Imaging

To characterzie the contrast uptake heterogeneity in perfusion MRI images, as a first step, a 4 dimensional representation is built for the kinetic curve for all the 3D voxels. The voxels are then grouped according to the timepoint at which peak enhancement was attained. Specifically, this step partitions the voxels into three groups: voxels with rapid, moderate, and slow enhancement. Once the voxels are partitioned into groups, within each group, first and second order statistics are computeed for enhancementto build a feature descriptor for the entire 3D image volume. This feature representation wasmotivated by the intention to capture the spatiotemporal heterogeneity in MRI images with respect to the speed of the contrast uptake. Using these features a hierarchical clustering approach is developed for identifying intrinsic imaging subtypes in breast cancer tumors. Specifically, the unsupervised clustering based on the heterogeneity encoding features automatically resulted in groupings of tumor images which were well separated in terms of their recurrence risk.

Ahmed Ashraf, S. Gavenonis, D. Daye, C. Mies, M. Feldman, M.A. Rosen, and D. Kontos, “(Unsupervised) Identification of intrinsic radio-phenotypes for breast cancer tumors: Preliminary associations with prognostic gene expression profiles.”, Radiology, 2014,(Impact Factor: 7.608) [Publication Interview]

Ahmed Ashraf, B. Gaonkar, C. Mies, A. DeMichele, M. Rosen, C. Davatzikos, D. Kontos, “Breast DCE-MRI Kinetic Heterogeneity Tumor Markers: Preliminary Associations With Neoadjuvant Chemotherapy Response”, Translational Oncology, 2015 [Link]


Multichannel Markov Random Fields (MRFs) for Image Segmentation from Perfusion MRI Imaging

In perfusion imaging, multiple features/properties can be computed for every voxel location e.g. time-to-peak, peak-enhancement, and a statistics capturing the characeristics of the contrast-uptake kinetic curve. To make use of multiple features for segmentation, a multi-channel Markov random field is presented which enables probabilistic inference at the level of each super-voxel subject to a smoothness objective.

A. Ashraf, S. Gavenonis, D. Daye, C. Mies, M. Feldman, M.A. Rosen, and D. Kontos, “A Multichannel Markov Random Field Framework for Tumor Segmentation with an Application to Classification of Gene Expression-based Breast Cancer Recurrence Risk”, IEEE Transactions on Medical Imaging (IEEE TMI), April 2013 [Link]


Learning to Unlearn: Building Immunity to Dataset Bias in Biomedical Imaging
   
 

In this project, we situate the problem of dataset bias in the context of medical imaging studies. We show empirical evidence that such a problem exists in medical datasets. We then present a framework to unlearn study membership as a means to handle the problem of database bias. Our main idea is to take the data from the original feature space to an intermediate space where the data points are indistinguishable in terms of which study they come from, while maintaining the recognition capability with respect to the variable of interest. This will promote models which learn the more general properties of the etiology under study instead of aligning to dataset-specific peculiarities. Essentially, our proposed model learns to unlearn the dataset bias.

A. Ashraf, S. Khan, N. Bhagwat, M. Chakravarty, B. Taati, "Learning to Unlearn: Building immunity to dataset bias in medical imaging studies", ML4H, NIPS, 2018 [Link]


Subspace Transfer Learning for Facial Landmark Detection
   
 

We propose a subspace transfer learning method, in which we select a subspace from the source that best describes the target space. We propose a metric to compute the directional similarity between the source eigenvectors and the target subspace. We show an equivalence between this metric and the variance of target data when projected onto soure eigenvectors. Using this equivalence, we select a subset of source principal directions that capture the variance in target data. To define our model, we augment the selected source subspace with the target subspace learned from a handful of target examples.

Azin Asgarian, Ahmed Ashraf, David Fleet, and Babak Taati, “Subspace selection to suppress confounding source domain information in AAM transfer learning”, IEEE International Conference on Biometrics (IEEE IJCB), 2017 [Link]


Computer Vision Based Analysis of Handwashing Behavior in Older Adults for Cognitive Health Assessment
   
 

In this project an algorithm was developed to detect the cognitive status of older adults through video tracking of a frequent activity of daily living (ADL) such as hand washing. A dataset was built by recording videos of older adults from an overhead camera as they washed their hands. The first step in this work involved the segmentation and localization of a person's hands. Features were then extracted capturing the location, speed, the path tortuosity (complexity) as measured by the fractal dimension of the hand trajectory, along with time spent during different steps while washing hands. A random forest classifier trained on these features was able to detect the cognitive status in terms of mild, moderate, and severe dementia.

Ahmed Ashraf and Babak Taati, “Automated video analysis of handwashing behavior as a potential marker of cognitive health in older adults”, IEEE Journal of Biomedical and Health Informatics (IEEE JBHI), 2016 [Link]


Pain Expression Recognition for Older Adults with Dementia living in Long Term Care
   
 

Babak Taati, Shun Zhao, Ahmed Ashraf, Azin Asgarian, M Erin Browne, Kenneth M Prkachin, Alex Mihailidis, Thomas Hadjistavropoulos, “Algorithmic Bias in Clinical Populations–Evaluating and Improving Facial Analysis Technology in Older Adults with Dementia”, IEEE Access, 2019 [Link]

M. Erin Browne, Thomas Hadjistavropoulos, Kenneth Prkachin, Ahmed Ashraf, Babak Taati, “Pain Expressions in Dementia: Validity of Observers’ Pain Judgments as a Function of Angle of Observation”, Journal of Non-verbal Behavior, 2019 [Link]

Ahmed Ashraf, S. Lucey, J. F. Cohn, T. Chen, K. M. Prkachin, and P. E. Solomonl, “The Painful Face – Pain Expression recognition Using Active Appearance Models”, International Journal of Image and Vision Computing, 2009 [Link]


Image Alignment and Tracking in Fourier Space
   
 

We have shown that the conventional Lucas-Kanade algorithm can equivalently be cast in the Fourier domain, instead of the conventional spatial domain. With this formulation, we have shown that performing alignment in the high dimensional feature spaces derived from a bank of filters becomes mathematically equivalent to performing alignment in the low dimensional image space, if appropriate weightings are applied in the Fourier domain. This technique renders the Fourier-LK algorithm robust to noisy artifacts like illumination in comparison to the conventional Lucas-Kanade algorihtm.

Ahmed Ashraf, Simon Lucey, Tsuhan Chen, “Fast Image Alignment in the Fourier domain”, IEEE International Conference on Computer Vision and Pattern recognition (IEEE CVPR) 2010 [Link]

Simon Lucey, Rajitha Navarathna, Ahmed Ashraf, Sridha Sridharan, “Fourier Lucas Kanade Algorithm”, IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), 2012 [Link]


Fourier SVMs and Filter Learning
   
   

Ahmed Ashraf, Simon Lucey, Tsuhan Chen , “Reinterpreting the application of Gabor filters as a manipulation of the margin in the Support Vector Machines”, IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI), 2010 [Link]

Simon Lucey, Ahmed Ashraf, “Nearest neighbor classifier generalization through spatially constrained filters”, Pattern Recognition, 2013 [Link]


 

Disclaimer

This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All person copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.