Home Research Medical Applications

 

 


Classification of microscopic images

In this project we focus on diagnosing pre-malignant changes in Barrett's Esophagus -- a disease of the junction between the stomach and the esophagus.  Over time, this disease may progress, first by transforming into a pre-malignant condition called low grade dysplasia*, then into a more severe stage called high grade dysplasia, and finally into adenocarcinoma (i.e. full cancer).

If a patient is diagnosed with Barrett’s esophagus with no dysplasia, then he/she will be submitted to a follow-up biopsy.  If a patient is diagnosed with low-grade dysplasia, a more severe follow-up biopsy will be applied, that is every 6 months.  And finally, if a patient is diagnosed with high grade dysplasia, especially in multiple biopsy sites, then the esophagectomy (surgical excision of the esophagus) is considered.

The aim of this project is to use computer vision techniques in order to develop a system that can classify different stages of dysplasia and predict progression to invasive adenocarcinoma in Barrett’s esophagus. Epithelial dysplasia in Barrett’s esophagus is the precursor of invasive adenocarcinoma and provides the basis for endoscopic surveillance and treatment decisions. A significant correlation exists between the degree of dysplasia and the time of progression to invasive adenocarcinoma.  However, since the current grading scheme of dysplasia is primarily based on the subjective evaluation, considerable interobserver variability in the grading of dysplasia has been encountered.  The main difficulty lies in discriminating between the no-dysplasia and low dysplasia stages, and between the low-dysplasia  and the high grade dysplasia. The overall agreement between two pathologist looking at the same tissue and deciding the grade of dysplasia is between 58% and 61%!  Given such high inter-observer variability and possibly disastrous implications of incorrect diagnosis, there is a great need for a system for a system for accurate and automatic classification of microscopic images.

Barrett’s no dysplasia

Low grade dysplasia

High grade dysplasia

Adenocarcinoma

* Dysplasia is an abnormality in the appearance of cells

 


 

 

Home Research Medical Applications