Browse Prior Art Database

A Multiple Instance Learning Based Self Learning Approach to Reduce Human Effort in Marking Regions of Interest in Imaging Applications

IP.com Disclosure Number: IPCOM000213368D
Original Publication Date: 2011-Dec-13
Included in the Prior Art Database: 2011-Dec-13
Document File: 2 page(s) / 78K

Publishing Venue

Siemens

Related People

Juergen Carstens: CONTACT

Abstract

In the field of medical imaging the automatic detection of certain abnormal anatomies is of interest for further processing in different modalities. If anomalies can be detected automatically, the throughput time of each patient can be reduced. Thus, more patients can be treated.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 2

A Multiple Instance Learning Based Self Learning Approach to Reduce Human Effort in Marking Regions of Interest in Imaging Applications

Idea: Chhaya Methani, IN-Bangalore; Rahul Thota, IN-Bangalore; Amit Kale, Ph.D., IN-Bangalore

In the field of medical imaging the automatic detection of certain abnormal anatomies is of interest for further processing in different modalities. If anomalies can be detected automatically, the throughput time of each patient can be reduced. Thus, more patients can be treated.

For this automation, a classifier needs to be learned. Typically, a large number of positive and negative examples have to be labeled by an expert in order to be used for this training. Here, a positive example contains the searched object while a negative example does not.

The labeling involves a lot of human effort and it is cost and time intensive. Thus, the objective is to use unlabeled data for the classifier's training. For this purpose, the classifier can be trained with a small amount of labeled data first before letting the classifier label the data coarsely and then refining the choices. More information can be added to the classifier by showing difficult examples of this labeling to an expert for annotation. This kind of approach to learning a classifier is called semi- supervised learning. For the expert this means considerably less effort in reviewing and labeling data than in the supervised learning approach.

In the concept of Multi Instance Learning (MIL), a classifier is trained not with a set of instances that are labeled positive or negative, but with a set of positive and negative bags. Each of these bags may contain many instances. A negative bag contains only negative...