Automated Detection of Major Thoracic Structures with a Novel Online Learning Method

Jianming Liang (Inventor)

Research output: Patent

Abstract

Automated detection of anatomic structures is an essential functionality in navigating large 3D image datasets and supporting computer-aided diagnosis (CAD). Several approaches have been proposed for localizing anatomic structures, but all of these methods are offline, require advance training, and offer no capability for continually improving their performance. Conventional offline learning requires collecting all representative samples before the commencing training. Researchers at Arizona State University have developed a novel online learning method for automatically detecting anatomic structures in medical images, which continually updates a linear classifier. Given a set of training samples, it dynamically updates a pool containing M features and returns a subset of N best features along with their corresponding voting weights. Compared to Grabner's approach, this method demonstrated superior performance for three thoracic structures of high diagnostic value (pulmonary trunk, carina, and aortic arch). In addition, the online aspect is a significant advantage because of the dynamic correction based on feedback from physicians. Potential Applications Anatomy detection Benefits and Advantages Continually updates a linear classifier Greater computational efficiency Higher resilience to outlier samples Better keeps the performance on the previous samples Dowload Original PDF For more information about the inventor(s) and their research, please see Dr. Liang's departmental webpage
Original languageEnglish (US)
StatePublished - Sep 16 2011

Fingerprint

Classifiers
Computer aided diagnosis
Arches
Computational efficiency
Feedback

Cite this

@misc{986bef8696144a9fb6675fb9ac7bc355,
title = "Automated Detection of Major Thoracic Structures with a Novel Online Learning Method",
abstract = "Automated detection of anatomic structures is an essential functionality in navigating large 3D image datasets and supporting computer-aided diagnosis (CAD). Several approaches have been proposed for localizing anatomic structures, but all of these methods are offline, require advance training, and offer no capability for continually improving their performance. Conventional offline learning requires collecting all representative samples before the commencing training. Researchers at Arizona State University have developed a novel online learning method for automatically detecting anatomic structures in medical images, which continually updates a linear classifier. Given a set of training samples, it dynamically updates a pool containing M features and returns a subset of N best features along with their corresponding voting weights. Compared to Grabner's approach, this method demonstrated superior performance for three thoracic structures of high diagnostic value (pulmonary trunk, carina, and aortic arch). In addition, the online aspect is a significant advantage because of the dynamic correction based on feedback from physicians. Potential Applications Anatomy detection Benefits and Advantages Continually updates a linear classifier Greater computational efficiency Higher resilience to outlier samples Better keeps the performance on the previous samples Dowload Original PDF For more information about the inventor(s) and their research, please see Dr. Liang's departmental webpage",
author = "Jianming Liang",
year = "2011",
month = "9",
day = "16",
language = "English (US)",
type = "Patent",

}

TY - PAT

T1 - Automated Detection of Major Thoracic Structures with a Novel Online Learning Method

AU - Liang, Jianming

PY - 2011/9/16

Y1 - 2011/9/16

N2 - Automated detection of anatomic structures is an essential functionality in navigating large 3D image datasets and supporting computer-aided diagnosis (CAD). Several approaches have been proposed for localizing anatomic structures, but all of these methods are offline, require advance training, and offer no capability for continually improving their performance. Conventional offline learning requires collecting all representative samples before the commencing training. Researchers at Arizona State University have developed a novel online learning method for automatically detecting anatomic structures in medical images, which continually updates a linear classifier. Given a set of training samples, it dynamically updates a pool containing M features and returns a subset of N best features along with their corresponding voting weights. Compared to Grabner's approach, this method demonstrated superior performance for three thoracic structures of high diagnostic value (pulmonary trunk, carina, and aortic arch). In addition, the online aspect is a significant advantage because of the dynamic correction based on feedback from physicians. Potential Applications Anatomy detection Benefits and Advantages Continually updates a linear classifier Greater computational efficiency Higher resilience to outlier samples Better keeps the performance on the previous samples Dowload Original PDF For more information about the inventor(s) and their research, please see Dr. Liang's departmental webpage

AB - Automated detection of anatomic structures is an essential functionality in navigating large 3D image datasets and supporting computer-aided diagnosis (CAD). Several approaches have been proposed for localizing anatomic structures, but all of these methods are offline, require advance training, and offer no capability for continually improving their performance. Conventional offline learning requires collecting all representative samples before the commencing training. Researchers at Arizona State University have developed a novel online learning method for automatically detecting anatomic structures in medical images, which continually updates a linear classifier. Given a set of training samples, it dynamically updates a pool containing M features and returns a subset of N best features along with their corresponding voting weights. Compared to Grabner's approach, this method demonstrated superior performance for three thoracic structures of high diagnostic value (pulmonary trunk, carina, and aortic arch). In addition, the online aspect is a significant advantage because of the dynamic correction based on feedback from physicians. Potential Applications Anatomy detection Benefits and Advantages Continually updates a linear classifier Greater computational efficiency Higher resilience to outlier samples Better keeps the performance on the previous samples Dowload Original PDF For more information about the inventor(s) and their research, please see Dr. Liang's departmental webpage

M3 - Patent

ER -