Evaluation of image features and classification methods for Barrett's cancer detection using VLE imaging

Sander R. Klomp, F. van der Sommen, A.-F. Swager, S. Zinger, E.J. Schoon, W.L. Curvers, J.J. Bergman, P.H.N. de With

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

11 Citaten (Scopus)
1 Downloads (Pure)

Samenvatting

Volumetric Laser Endomicroscopy (VLE) is a promising technique for the detection of early dysplasia in Barrett's Esophagus (BE). VLE generates hundreds of high-resolution, grayscale, cross-sectional images of the esophagus.
However, at present, classifying these images is a time consuming and cumbersome eort performed by an expert using a clinical prediction model. This paper explores the feasibility of using computer vision techniques to accurately predict the presence of dysplastic tissue in VLE BE images with known histology. Our contribution is twofold. First, a benchmarking is performed for widely applied machine learning techniques and feature extraction methods. Second, three new features based on the clinical detection model are proposed, with superior classication accuracy and execution speed, compared to earlier work. The results are evaluated on a clinically validated dataset of 30 dysplastic and 30 non-dysplastic VLE images with known histology. Optimal classication accuracy is obtained by applying Adaptive Boosting with decision trees and using our modied Haralick features, yielding an area under the receiver operating characteristic of 0.91 compared to the clinical prediction model at 0.81. Optimal execution time is achieved using a proposed mean and median feature, which is extracted at least factor 2.5 faster than features with comparable performance.
Originele taal-2Engels
TitelMedical Imaging 2017 : Computer-Aided Diagnosis, February 11-16, 2017, Orlando, Florida, USA
RedacteurenS.G. Armato, N.A. Petrick
UitgeverijSPIE
Aantal pagina's10
ISBN van elektronische versie9781510607132
ISBN van geprinte versie978-1-5106-0713-2
DOI's
StatusGepubliceerd - 2017
EvenementSPIE Medical Imaging: Image Processing - Renaissance Orlando at Sea World/Orlando, Orlando, Verenigde Staten van Amerika
Duur: 11 feb 201716 feb 2017

Publicatie series

NaamProceedings of SPIE
Volume10134

Congres

CongresSPIE Medical Imaging
LandVerenigde Staten van Amerika
StadOrlando
Periode11/02/1716/02/17

Vingerafdruk Duik in de onderzoeksthema's van 'Evaluation of image features and classification methods for Barrett's cancer detection using VLE imaging'. Samen vormen ze een unieke vingerafdruk.

Citeer dit