Detection and analysis of branching structures and/or texture is very challenging; it arises in many areas of science and engineering (e.g., medical images, chemical compounds, etc). The objective of this proposal is to develop novel approaches to model, detect, and analyze branching structures obtained from multimodality data. Such representation and analysis tools are expected to make many complex problems more tractable. Examples include identifying and recognizing a large number of structure classes; discovering new relationships among structure, texture, and function or pathology; evaluating hypotheses; developing modeling tools; assisting with surgical design; and managing medical image data efficiently. Specifically, the investigators plan to explore three research topics under this project: (1) To develop descriptors of branching structures and texture, and knowledge discovery tools that will enable hypotheses generation and evaluation and improve modeling of branching structures; (2) To design automated algorithms and a flexible framework to detect branching structures. The investigators are especially interested in addressing challenges of occlusion and topology change; (3) To demonstrate the applicability of the proposed tools to breast imaging by building a prototype database of images from various modalities and associated clinical data that will provide advanced analysis and visualization capabilities. Though the investigators use breast imaging as the driving application, the proposed project is expected to provide software and data resources that can assist clinical tasks and scientific discoveries in general. Developing automated tools to effectively characterize, detect, and classify tree-like structures in images would provide great insight into the relationship between the branching topology and function or pathology. The investigators plan to further contribute to the medical/scientific community by disseminating the related software and annotated data sets. The educational goals include incorporating research findings to graduate courses at Temple (data mining course and medical image analysis seminar) and at the University of Pennsylvania (medical image analysis course).
This Collaborative Research Grant has been focused on modeling, detection, and analysis of branching structures in medical images. Branching structures frequently occur in human anatomy, and are usually related to hierarchical organization, e.g., hierarchy of blood vessels, bronchial network of the lungs, breast ductal network, etc. Branching properties of these structures may indicate the presence or an increased risk of disease. The branching?structures are clinically visualized using various imaging modalities, both directly?and indirectly through?the contribution of these structures to background image texture. This project is related to a long-term research focus of our X-ray Physics Laboratory at the University of Pennsylvania, on the computer simulation of breast anatomy and clinical imaging process,?for the purpose of breast imaging validation. Our lab has extensive experience in developing computer models of breast anatomy (also called breast phantoms), currently used by over 50 research?groups from 16 countries in academia, industry, and government. Images synthesized using software phantoms can?be used for preclicnial validation and optimization of breast imaging system, thus reducing the cost, duration and the need for repeated irradiation of volunteer patients during clinical imaging trials. Better understanding of breast ductal branching properties and their appearance in clinical images, as provided by this grant, helps improve the realism of simulated anatomy and synthetic images,?increasing the quality of preclinical validation of breast imaging. In addition to cancer detection, breast imaging has been used incresingly for the estimation of cancer risk.? Image-based risk biomarkers (breast density and parenchymal features) are added to personal, family and hormonal risk factors, to better identify candidates for risk-modifying drug therapy. Our previous analysis of ductal branching properties as visualized by galactography (contrast enhanced x-ray imaging of breast ducts) suggests correlation betwen cancer risk and ductal network topology; this correlation is also supported by the evidence from mouse cancer models. Clinical visualization of breast ducts is, however, not routinely performed; galactography is rarely indicated and reveals mostly benign findings. On the other hand, mammographic texture (i.e., parenchymal pattern) has been long investigated for the correlation with risk. As a part of this grant we tested hypothesis about the correlation between?texture and topological descriptors. To that end, we selected descriptors of parenchymal texture among those used for cancer risk estimation.? Texture features were calculated using an automated breast image analysis pipeline, developed at the University of Pennsylvania. For ductal topology, we used descriptors previously validated in classification between normal, benign, and malignant galactograms. These descriptors were developed by our collaborating partner on this NSF grant from Temple University. Hypothesis testing was performed using anonymized, previoulsy acquired clinical mammograms and galactograms from 49 women, provided to us from Virginia Commonwealth University. Clinical images were digitized from film, and categorized based upon the visibility of ductal network. After manually tracing ductal networks,?a subset of 41 galactograms?from 13 patients with good ductal visibility was selected for testing. We also calculated 26 different texture features from 56 corresponding mammograms. We have calculated regression between the texture features (averaged over all mammograms of the same patient) and the topological properties calculated from the corresponding traced ductal networks (averaged over all galactograms of the same patient). Initial analysis could not achieve statistical significance, due to a small sample size. To assess the effect of samepl size, we simulated an increased dataset by bootstrapping. The analysis of such increased dataset suggested statistically significant regression between topological descriptors and a number of texture features. We will continue with research activities after the end of grant funding period. We plan to increase the dataset by adding more clinical images. (We expect to double the sameple size in the near future. The increased dataset, along with corresponding database of topological and texture descriptors will be made publicly available.) The analysis of such increased dataset will verify the findings suggested from our analysis of bootstrap data, anticipating teh following outcomes: If the correlation between the texture and topology is confirmed (as suggested by the bootstrapped analysis), texture descriptors could be used as proxy for topology, as the latter is not routinely visible in clinical images; Identifying texture feature(s) or their combination, with strongest correlation with topology, would improve the understanding of texture biomarkers of cancer risk; and potentially increase the ?accuracy of cancer risk estimation; Improved understanding of topological descriptors and the topology-texture correlation would also help increase the realism of computer simulation of breast anatomy and imaging; Lastly, if, however, the increased sample size does not confirm the correlation between topology and texture, that could suggest that the topology may carry the risk-related information, which is independent from texture descriptors.? This could lead to potentially improved risk estimation, providing a clinically feasible method?for the visualization of breast ducts.
