Papers by Tejaswini Ganapathi
Proceedings of SPIE, Feb 23, 2012
The goal of this study was to assess if stereoscopic viewing of breast tomosynthesis projection i... more The goal of this study was to assess if stereoscopic viewing of breast tomosynthesis projection images impacted mass detection performance when compared to monoscopic viewing. The dataset for this study, provided by Hologic, Inc., contained 47 craniocaudal cases (23 biopsy proven malignant masses and 24 normals). Two projection images that were separated by 8 degrees were chosen to form a stereoscopic pair. The images were preprocessed to enhance their contrast and were presented on a stereoscopic display. Three experienced breast imagers participated in a blinded observer study as readers. Each case was shown twice to each reader -once in the stereoscopic mode, and once in the monoscopic mode in a random order. The readers were asked to make a binary decision on whether they saw a mass for which they would initiate a diagnostic workup or not, and also report the location of the mass and provide a confidence score in the range of 0-100. The binary decisions were analyzed using the sensitivity-specificity measure, while the confidence scores were analyzed using the Receiver Operating Characteristic curve (ROC). We also report a statistical analysis of the difference in partial AUC values greater than 95% sensitivity between the stereoscopic and monoscopic modes.
In Dictionary Learning one tries to recover incoherent matrices A * ∈ R n×h (typically overcomple... more In Dictionary Learning one tries to recover incoherent matrices A * ∈ R n×h (typically overcomplete and whose columns are assumed to be normalized) and sparse vectors x * ∈ R h with a small support of size h p for some 0 < p < 1 while having access to observations y ∈ R n where y = A * x * . In this work we undertake a rigorous analysis of whether gradient descent on the squared loss of an autoencoder can solve the dictionary learning problem. The Autoencoder architecture we consider is a R n → R n mapping with a single ReLU activation layer of size h. Under very mild distributional assumptions on x * , we prove that the norm of the expected gradient of the standard squared loss function is asymptotically (in sparse code dimension) negligible for all points in a small neighborhood of A * . This is supported with experimental evidence using synthetic data. We also conduct experiments to suggest that A * is a local minimum. Along the way we prove that a layer of ReLU gates can be set up to automatically recover the support of the sparse codes. This property holds independent of the loss function. We believe that it could be of independent interest.
arXiv (Cornell University), Jul 23, 2016
The purpose of this research is to evaluate human and technical factors required to create a huma... more The purpose of this research is to evaluate human and technical factors required to create a human-computer interface (HCI) for a radiology structured reporting process that incorporates eye-gaze and speech signals. Gaze and speech signals generated by radiologists during simulated image interpretation and dictation tasks were analyzed to describe the a) variation of gaze -speech temporal relationships in a radiology workflow environment, and b) variation in eye movements for a particular image interpretation task among different radiologists. Knowledge of these factors provides information regarding the complexity of the image interpretation and dictation task which can be used to design a HCI for future use in diagnostic radiology. Our goal is to use these data to create an HCI to automate the generation of a particular type of structured radiology report that captures key image findings and a radiologist's subsequent spoken descriptions of those findings to create a multimedia report . This study received Institutional Review Board approval. The radiologic images used in this study were de-identified to comply with HIPAA requirements. Five radiologists performed image interpretation and dictation tasks on 30 images from 3 modalities (10 images per modality), during which their eye gaze and speech signals were recorded using a Tobii eye-tracking device (Tobii T60 XL, Tobii Technology, Danderyd, Sweden). The dictation tasks were designed to be consistent across each modality and with each radiologist identifying two distinct targeted findings (e.g., identify and describe a lung mass on a chest x-ray followed by commenting on the heart size). Statistical modeling was performed to evaluate the variation of the eye gaze -speech temporal relationship on certain human and technical factors, such as the radiologist, image content, modality and resolution, and target order. Our results indicate that there are substantial differences in the gaze-speech relationship among radiologists (p<0.001). There is no statistically significant evidence that the gazespeech temporal relationship depends on the particular image modality (p=0.909) or target order (p=0.142). Analysis of gaze paths suggests that the search path variance among radiologists is significant. Our data indicate that a) gaze-speech relationships and b) scan paths vary significantly among radiologists, thus suggesting that a gaze-speech system for automating the capture of data for structured reporting processes may need to be customized for each user. The image resolution and layout, image content, and order of targets during an image interpretation session are irrelevant factors to consider when designing an HCI. Our findings can be applied to the design of other HCI solutions for radiological applications that involve the capture of visual data and human descriptions of image findings.
Traditional appearance based face recognition (FR) systems use gray scale images, however recentl... more Traditional appearance based face recognition (FR) systems use gray scale images, however recently attention has been drawn to the use of color images. Color inputs have a larger dimensionality, which increases the computational cost, and makes the small sample size (SSS) problem in supervised FR systems more challenging. It is therefore important to determine the scenarios in which usage of color information helps the FR system. In this thesis, it was found that inclusion of chromatic information in FR systems is shown to be particularly advantageous in poor illumination conditions. In supervised systems, a color input of optimal dimensionality would improve the FR performance under SSS conditions. A fusion of decisions from individual spectral planes also helps in the SSS scenario. Finally, chromatic information is integrated into a supervised ensemble learner to address pose and illumination variations. This framework significantly boosts FR performance under a range of learning ...
2018 IEEE International Symposium on Information Theory (ISIT), 2018
In Dictionary Learning one tries to recover incoherent matrices A * ∈ R n×h (typically overcomple... more In Dictionary Learning one tries to recover incoherent matrices A * ∈ R n×h (typically overcomplete and whose columns are assumed to be normalized) and sparse vectors x * ∈ R h with a small support of size h p for some 0 < p < 1 while having access to observations y ∈ R n where y = A * x * . In this work we undertake a rigorous analysis of whether gradient descent on the squared loss of an autoencoder can solve the dictionary learning problem. The Autoencoder architecture we consider is a R n → R n mapping with a single ReLU activation layer of size h. Under very mild distributional assumptions on x * , we prove that the norm of the expected gradient of the standard squared loss function is asymptotically (in sparse code dimension) negligible for all points in a small neighborhood of A * . This is supported with experimental evidence using synthetic data. We also conduct experiments to suggest that A * is a local minimum. Along the way we prove that a layer of ReLU gates can be set up to automatically recover the support of the sparse codes. This property holds independent of the loss function. We believe that it could be of independent interest.
ArXiv, 2016
The purpose of this research is to evaluate the human and technical factors required to create a ... more The purpose of this research is to evaluate the human and technical factors required to create a human-computer interface (HCI) for a structured reporting solution based on eye-gaze and speech signals. Gaze and speech signals from radiologists acquired during simulated image interpretation and dictation sessions were analyzed to determine a) variation of temporal relationship between eye gaze and speech in a dictation environment, and b) variation in eye movements for a particular image interpretation task among radiologists. Knowledge of these factors provides information regarding the complexity of the image interpretation or dictation task, and provides information that can be used to design a HCI for use in diagnostic radiology. Our ultimate goal is to use these data to create an HCI to automate the generation of a particular type of structured radiology report. Our data indicate that the a) temporal relationships between eye gaze and speech and b) scan paths substantially vary ...
A Vision for Radiology Structured Reporting
ABSTRACT
Use of Eye-Tracking Technology and Natural Language Processing for Data Input into a Multimedia Structured Report
ABSTRACT
Traditional appearance based face recognition (FR) systems use gray scale images, however recentl... more Traditional appearance based face recognition (FR) systems use gray scale images, however recently attention has been drawn to the use of color images. Color inputs have a larger dimensionality, which increases the computational cost, and makes the small sample size (SSS) problem in supervised FR systems more challenging. It is therefore important to determine the scenarios in which usage of color information helps the FR system. In this thesis, it was found that inclusion of chromatic information in FR systems is shown to be particularly advantageous in poor illumination conditions. In supervised systems, a color input of optimal dimensionality would improve the FR performance under SSS conditions. A fusion of decisions from individual spectral planes also helps in the SSS scenario. Finally, chromatic information is integrated into a supervised ensemble learner to address pose and illumination variations. This framework significantly boosts FR performance under a range of learning scenarios.
SPIE Proceedings, 2012
The goal of this study was to assess if stereoscopic viewing of breast tomosynthesis projection i... more The goal of this study was to assess if stereoscopic viewing of breast tomosynthesis projection images impacted mass detection performance when compared to monoscopic viewing. The dataset for this study, provided by Hologic, Inc., contained 47 craniocaudal cases (23 biopsy proven malignant masses and 24 normals). Two projection images that were separated by 8 degrees were chosen to form a stereoscopic pair. The images were preprocessed to enhance their contrast and were presented on a stereoscopic display. Three experienced breast imagers participated in a blinded observer study as readers. Each case was shown twice to each reader -once in the stereoscopic mode, and once in the monoscopic mode in a random order. The readers were asked to make a binary decision on whether they saw a mass for which they would initiate a diagnostic workup or not, and also report the location of the mass and provide a confidence score in the range of 0-100. The binary decisions were analyzed using the sensitivity-specificity measure, while the confidence scores were analyzed using the Receiver Operating Characteristic curve (ROC). We also report a statistical analysis of the difference in partial AUC values greater than 95% sensitivity between the stereoscopic and monoscopic modes.
Uploads
Papers by Tejaswini Ganapathi