Browse Prior Art Database

Incrementally Learning Sample-Specific Ensemble of Segmentation Techniques in Medical Imaging

IP.com Disclosure Number: IPCOM000244845D
Publication Date: 2016-Jan-21
Document File: 5 page(s) / 107K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a process that combines the outputs from several segmentation processes by considering individual image data (on which the objects of interests are to be located). The process is applicable to any general image processing problem.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 47% of the total text.

Page 01 of 5

Incrementally Learning Sample - Medical Imaging

Medical examination and clinical practice routinely use medical images , such as Computed Axial Tomography (CAT), Magnetic Resonance Imaging (MRI), Ultrasound, X-Ray, and dermoscopy images. Research efforts have been devoted to processing and analyzing medical images to extract meaningful information . Automated image segmentation, which aims at automated extraction of object boundary features , plays a fundamental role in understanding image content for searching and mining in medical image archives.

The first step of any automated (or semi-automated) image processing task is to identify the objects of interest in the image. This is called the segmentation process. Segmentation results in defining a boundary around the objects of interest . Relevant features (e.g., size, shape, texture, marks/signatures, etc.) are then extracted from these areas to serve a multitude of applications (e.g., diagnosis, statistics, understanding, modelling, etc.). The success of segmentation is crucial for a useful image processing task.

Almost every existing segmentation technique relies on having fixed or pre-defined steps or operations performed on images of certain types (e.g., RGB, Hyperspectral, etc.) and sources (e.g., dermoscopy, microscopy, standard camera etc.).

Performances of different segmentation techniques vary across different images depending on a number of factors including image quality , image source, present in the image, etc. Combining different segmentation techniques is a promising avenue to attain higher accuracy. However, existing literature presents combination methods that use pre-defined and hard-coded fusion weights.

The core motivation, then, is to:

1. Incorporate a sample-specific late fusion mechanism to dynamically generate fusion weight based for individual samples

2. Implement an incrementally learning mechanism to evolve the performance of the fusion as more training samples are provided

The novel contribution is a process that combines the outputs from several segmentation processes by considering individual image data (on which the objects of interests are to be located). The process is applicable to any general image processing problem.

The disclosed system and method apply a sample -specific late fusion of various border detection techniques to find the optimal border in a given medical image . The method incrementally evolves to improve the overall performance as more training samples are provided. The motivation of the proposed algorithm is based on the fact that , depending on the inherit nature and certain visual characteristics of each individual sample image , one segmentation technique might be more effective than others . Thus, the proposed

1

-Specific Ensemble of Segmentation Techniques in

Specific Ensemble of Segmentation Techniques in


Page 02 of 5

algorithm generates output by combining multiple segmentation results of different methods. In this techn...