ARTIFICIAL STAINING OF HISTOPATHALOGICAL IMAGES
Principal Investigator
Name
Alper Yilmaz
Degrees
Ph.D
Institution
The Ohio State University
Position Title
Professor
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-516
Initial CDAS Request Approval
May 31, 2019
Title
ARTIFICIAL STAINING OF HISTOPATHALOGICAL IMAGES
Summary
Immunohistochemical techniques plays a critical role in screening, diagnosis and identification of neuroendocrine markers for lung cancer. Examining Hematoxylin and Eosin (H&E) tissue samples are often the initial steps taken towards the fight of a cancer. A patient’s treatment and possible victory in this fight ultimately depends on the accurate initial identification and characterization of the tumor cells. Although the pathologists are highly trained individuals in identifying abnormal cell growths, when dealing with images in the pixel level, the error for missing or misclassifying could have devastating consequences.
To aid in the accurate diagnosis of pathological images, we propose to research deep learning techniques that will allow for accurate artificial staining of tissue samples which in turn will result in accurate cancer diagnostics. Many of the immunohistochemical analyses are affected by pre-analytical handling of the tissue samples. The roots of these handling issues could be traced back to human error, with the help of artificial staining, the errors that occur during handling can be mitigated. Along with the reduction of handling errors, deep learning based artificial staining could help speed up the immunohistochemical staining process. Accurate and fast results can help reduce the man power needed for manual staining of samples, and instead, allow those individuals to focus their efforts on improving other aspects of the treatment process.
Many deep learning techniques are available to handle complex multi-class classification problems, which have been proven extremely effective in medical and non-medical domains. The foundation of most image based deep learning solutions is the Convolutional Neural Network (CNN). Although traditional CNN’s are extremely effective, many techniques have been developed to augment the CNN, which produce fast and accurate results. The researchers at The Ohio State University are interested in the comparisons between the application of two State-Of-the Art techniques in computer vision on pathological images. Generative Adversarial Networks (GAN’s) and Region Proposal Networks are two influential techniques in the Computer Vision field that have been proven fruitful. We are interested in the determining whether one technique individually or the combination of the two would result in accurate artificial staining of tissue specimens.
To aid in the accurate diagnosis of pathological images, we propose to research deep learning techniques that will allow for accurate artificial staining of tissue samples which in turn will result in accurate cancer diagnostics. Many of the immunohistochemical analyses are affected by pre-analytical handling of the tissue samples. The roots of these handling issues could be traced back to human error, with the help of artificial staining, the errors that occur during handling can be mitigated. Along with the reduction of handling errors, deep learning based artificial staining could help speed up the immunohistochemical staining process. Accurate and fast results can help reduce the man power needed for manual staining of samples, and instead, allow those individuals to focus their efforts on improving other aspects of the treatment process.
Many deep learning techniques are available to handle complex multi-class classification problems, which have been proven extremely effective in medical and non-medical domains. The foundation of most image based deep learning solutions is the Convolutional Neural Network (CNN). Although traditional CNN’s are extremely effective, many techniques have been developed to augment the CNN, which produce fast and accurate results. The researchers at The Ohio State University are interested in the comparisons between the application of two State-Of-the Art techniques in computer vision on pathological images. Generative Adversarial Networks (GAN’s) and Region Proposal Networks are two influential techniques in the Computer Vision field that have been proven fruitful. We are interested in the determining whether one technique individually or the combination of the two would result in accurate artificial staining of tissue specimens.
Aims
Develop a deep learning network to artificially stain tissue samples.
Collaborators
Shehan Perera, Graduate Student, The Ohio State University - Photogrammetric Computer Vision Lab