User Tools

Site Tools



This shows you the differences between two versions of the page.

Link to this comparison view

Next revision
Previous revision
projects:year5:16.07 [2019/08/21 09:31]
sally.johnson created
projects:year5:16.07 [2019/08/22 10:49] (current)
sally.johnson [Project - Documents]
Line 2: Line 2:
 ===== Project - Summary ===== ===== Project - Summary =====
 <WRAP leftalign box > <WRAP leftalign box >
-The project aims to develop and evaluate an incremental learning framework that can process high volumes of data without the need for retraining. The objectives are to: +To develop a general biomedical image informatics framework that is capable of predicting the traits of tissues captured in images of immunohistochemistry-stained (IHC) specimens. Our work on the framework has been on-going for a number of years. The CVDI Year 5 funding supported three separate sub-projects that contributed to the enhancement of the framework. They are
-  * develop multiple incremental learning algorithms (including semi-supervised and unsupervised+  * Definition of an alternate color representation for the analysis of Hematoxylin and Eosin (H&Eimages (a type of IHC image), which supports color normalization that effectively reduces inter-slide variability during downstream analysis. 
-  * evaluate the algorithms using simulated and real-world datasets +  * Estimation of fine-scale histologic features at low magnification which allows for the analysis of IHC images at lower pixel resolutions; thus providing computational speed-ups that facilitate high-throughput analysis of while-slide IHC images. 
-  * develop a customized incremental learner for author name disambiguation problem with IAB member, Clarivate Analytics+  * Development of an approach for characterization spatial structures in biomedical images based on N-point statistics analysis. The features derived from the analysis may be used to predict tissue traits.
 </WRAP> </WRAP>
 ===== Project - Team ===== ===== Project - Team =====
-^ Team Member     ^ Role              ^ Email          ^ Phone Number   ^ Academic Site/IAB  +^ Team Member         ^ Role                  ^ Email          ^ Phone Number   ^ Academic Site/IAB                  
-Gail Rosen      | PI                | Not available  | Not available  | Drexel University +David Breen         | PI                    | Not available  | Not available  | Drexel University                  | 
-Zhengqiao Zhao  Graduate Student  | Not available  | Not available  | Drexel University  |+| Mark Zarella        | PI                    | Not available  | Not available  | Drexel University                  
 +Chan Yeoh           | Student               | Not available  | Not available  | Drexel University                  | 
 +| Matthew Quaschnick  | Student               | Not available  | Not available  | Drexel University                  | 
 +| Dan Johnson         | Student               | Not available  | Not available  | Drexel University                  | 
 +| Zahra Riahi Samani  | Visiting PhD Student  | Not available  | Not available  | Shahid Beheshti University (Iran)  |
Line 18: Line 22:
 ===== Project - Impact and Uses/Benefits ===== ===== Project - Impact and Uses/Benefits =====
 <WRAP leftalign box > <WRAP leftalign box >
-The contributions of our research are thus three-fold(1) We propose four incremental learning algorithms that can update the decision boundary for the new incoming samples of the classifier without retraining the classifier with the entire training set each time. Moreoverone of the methods, IPAP, can also update the decision for original data (2) We proposed customized incremental learning algorithm (Incremental High precision rule base Naive Bayesian Classifier) to update bibliographic datasets without the need for labeled information. The disambiguation results from an existing disambiguation system as also optional in our framework. The high-precision rule NBC algorithm demonstrates high precision on a large-scale name block comprising nearly 20,000 citation records. To our knowledge, this test dataset is one of the largest name blocks ever used in an incremental author disambiguation evaluation research project(3) We successfully benchmark the performance of our proposed methods against other incremental approacheswhere we demonstrate notable improvements.+We envision multiple impacts and uses for the technologies developed in the three sub-projects. They are: 
 +  * SP1: Providing an improved tissue structure segmentation algorithms based on our alternative color space for H&E images. 
 +  * SP1: Supporting color normalizataion that will alleviate the problem of staining variability, a major impediment to automated processing and anaylsis of H&E images. 
 +  * SP2: The ability to estimate the histological composition of tissue in low-resolution images will facilitate the computational analysis of whole-slide images; thus enabling automated and high-throughput tissue analysis in pathology. 
 +  * SP3: N-point statistics provides a method for characterizing the architecture of tumors; thus improving the automated analysis/grading of tissues that should lead to improvedmore accurate and robust diagnosis of cancer.
 </WRAP> </WRAP>
Line 26: Line 34:
 ===== Project - Documents ===== ===== Project - Documents =====
projects/year5/16.07.1566397912.txt.gz · Last modified: 2019/08/21 09:31 by sally.johnson