Browse Prior Art Database

System and methods of risk estimation of retinal pathologies

IP.com Disclosure Number: IPCOM000248269D
Publication Date: 2016-Nov-14
Document File: 5 page(s) / 158K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed are a system and methods for one-to-one monitoring of retinal lesions, estimation of lesion growth rate, and prediction of the growth pattern to estimate the risk of disease.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 36% of the total text.

Page 01 of 5

System and methods of risk estimation of retinal pathologies

Measuring changes in the progression of retinal lesions in fundus images is important in disease management such as diabetic retinopathy (DR) and age related macular degeneration. For example, following a DR diagnosis and treatment plan implementation, patients should return for a follow-up evaluation approximately one year to every two-to-three months, depending on the severity of the disease. From a disease management perspective, it is not only important to track and monitor individual lesions, but also estimate the growth rate of the lesion in order to predict the future growth so that proper action can be taken if the progression is accelerating. It is time consuming and cumbersome for the doctors to track all the lesions and predict the growth of the lesions.

Very small lesions such as micro-aneurysm (in case of diabetic retinopathy) should be managed by close monitoring and, if the lesion size increases, the intervention has to be considered. For example, in case of diabetic retinopathy, the intervention may include medication to reduce blood glucose level. For other retinal lesions, radiotherapy or surgical excision may also be considered.

The novel contribution is a system and methods for one-to-one monitoring of retinal lesions, estimation of lesion growth rate, and prediction of the growth pattern to estimate the risk of the disease.

The system estimates the growth pattern of retinal pathologies of the lesions for risk assessment. The solution comprises the following methods:


• Segment optic disc and fovea in a retinal fundus image. A novel end-to-end deep learning based method to segment the optic disc and fovea by incorporating the distance constraint between the two structures.


• Calibrate retinal fundus images. The segmentation of optics disc (OD) and fovea is used to determine the parameter that maps the pixel measurement of lesions (e.g., width) into the real world unit such as millimeters (mm).

• Index retinal pathologies with respect to optic disc and the fovea
• Associate pathologies from a previous visit to the pathologies in the current visit


• Analyze the changes in pathologies across visits and estimate the growth rate


• Estimate the risk of the disease based on the growth rate and proximity to fovea and optic disc. For example, in age related macular degeneration, the size of drusens and its location with respect to fovea centralis is important to determine the severity of the disease.

Figure 1: System overview of tracking of retinal lesion progression

1


Page 02 of 5

Optic Disc, Fovea Segmentation

A deep learning architecture automatically segments the optic disc and fovea from the retinal images. The novel architecture consists of the following subparts:


• Deep fully convolutional neural network (FCNN) is a traditional FCNN block that is trained to perform pixel wise classification of the optic disc and fovea


• The deep feature i...