Brain tumors are the second- and fifth-most common cause of cancer death in males and females under 40, respectively. The 5-year relative survival rate is only 33%, even after decades of research into new treatments. Imaging based on """"""""virtual biopsy"""""""" can provide information about the entire lesion in a minimally or non-invasive way. Prior work by our group has demonstrated that image processing methods applied to serial examinations together (as opposed to applying them to individual examinations) can make subtle changes in brain tumors more apparent, allowing earlier detection of progression. We see the union of virtual biopsy methods and change detection methods as an innovative and powerful combination that our team is uniquely qualified to develop and evaluate. In this application, we will implement several feature selection (FS) methods in a tool that is """"""""image friendly"""""""". This tool will help select informative features from images;apply a wide range of existing ML algorithms to determine which features are important predictors of therapy response, and to evaluate the impact of a decision support application using these features and ML in a clinical trial. The final stage of th proposal will test the decision support tools in 3 clinical scenarios to see if they are able to significantly improve the decision making of clinicians in a clinical trial. The long-term goal of this proposal is to develop virtual biopsy technology that will enhance the clinical decision making process by providing tools for investigation of image-based therapy response assessment tools, that may also have some ability to predict outcome. We hope to apply the technology to other organ systems and other imaging technologies. We anticipate this project will impact clinical trials by enabling investigation of alternative outcome measures that are objectively assessed using algorithm evaluation methods. Such a toolset should be useful to the entire cancer imaging community to help evaluate features in old and new imaging technologies that correlate with patient survival. As such, this is ideal for helping the Quantitative Imaging Network (QIN) achieve its goals.
The selection of features that reflect response to therapy has long been the domain of the clinical radiologist. When imaging was relatively straightforward (e.g. a chest X-ray of lung cancer), a measurement or a visual assessment was a reasonable metric. Today, advanced imaging devices produce large amounts of data that reflect a range of properties. In this work, we build on previous work developing computer techniques for identifying and characterizing changes in brain tumors, using MRI. In this proposal, we will focus on building 1) a high quality database of brain cancer imaging, clinical data, """"""""-omic"""""""" data, outcomes data;2) a library of easily accessed tools for computing features that might be important predictors of tumor response;3) a tool that will help objectively establish the feature() that are valuable through a variety of machine learning methods;and 4) use the above to create a decision support tool that will be used in 3 clinical trial situations. .
|Kline, Timothy L; Edwards, Marie E; Garg, Ishan et al. (2018) Quantitative MRI of kidneys in renal disease. Abdom Radiol (NY) 43:629-638|
|Erickson, Bradley J; Korfiatis, Panagiotis; Kline, Timothy L et al. (2018) Deep Learning in Radiology: Does One Size Fit All? J Am Coll Radiol 15:521-526|
|Bae, Youngoh; Kumarasamy, Kunaraj; Ali, Issa M et al. (2018) Differences Between Schizophrenic and Normal Subjects Using Network Properties from fMRI. J Digit Imaging 31:252-261|
|Akkus, Zeynettin; Galimzianova, Alfiia; Hoogi, Assaf et al. (2017) Deep Learning for Brain MRI Segmentation: State of the Art and Future Directions. J Digit Imaging 30:449-459|
|Erickson, Bradley J; Korfiatis, Panagiotis; Akkus, Zeynettin et al. (2017) Toolkits and Libraries for Deep Learning. J Digit Imaging 30:400-405|
|Korfiatis, Panagiotis; Kline, Timothy L; Lachance, Daniel H et al. (2017) Residual Deep Convolutional Neural Network Predicts MGMT Methylation Status. J Digit Imaging 30:622-628|
|Kline, Timothy L; Korfiatis, Panagiotis; Edwards, Marie E et al. (2017) Image texture features predict renal function decline in patients with autosomal dominant polycystic kidney disease. Kidney Int 92:1206-1216|
|Erickson, Bradley J; Korfiatis, Panagiotis; Akkus, Zeynettin et al. (2017) Machine Learning for Medical Imaging. Radiographics 37:505-515|
|Peter, Roman; Korfiatis, Panagiotis; Blezek, Daniel et al. (2017) A quantitative symmetry-based analysis of hyperacute ischemic stroke lesions in noncontrast computed tomography. Med Phys 44:192-199|
|Hu, Leland S; Ning, Shuluo; Eschbacher, Jennifer M et al. (2017) Radiogenomics to characterize regional genetic heterogeneity in glioblastoma. Neuro Oncol 19:128-137|
Showing the most recent 10 out of 24 publications