Building a reduced dictionary of relevant perfusion patterns from ceus data for the classification of testis lesions

TY - CHAP

T1 - Building a reduced dictionary of relevant perfusion patterns from ceus data for the classification of testis lesions

AU - Favaron, Tommaso

AU - Huang, Dean Y.

AU - Christensen-Jeffries, Kirsten

AU - Eckersley, Robert E.

AU - Sidhu, Paul S.

AU - Grisan, Enrico

PY - 2019/4/1

Y1 - 2019/4/1

N2 - Radical orchifunicolectomy has traditionally been the main clinical treatment for small testicular masses (STMs); however STMs represent a constantly increasing and often incidental finding. Since many of them result benign, a more conservative testis-sparing surgery was proposed, but it requires a preliminary differentiation between benign and malignant masses: this however remains challenging. Although common understanding in radiology and oncology is that perfusion patterns might provide a useful information about the type of masses, no guidelines or consensus is available for the differentiation of STMs. We propose to build a dictionary of relevant perfusion patterns, extracted using non-negative matrix factorization on pixel-wise time-intensity curves from contrast-enhanced ultrasound data. When data from a lesion are reconstructed using this dictionary, a vector containing the frequency of utilization of each pattern can be used as a tissue signature. Using this signature, a support vector machine classifier has been trained, and the cross validated accuracy reached 100% in our pilot cohort.

AB - Radical orchifunicolectomy has traditionally been the main clinical treatment for small testicular masses (STMs); however STMs represent a constantly increasing and often incidental finding. Since many of them result benign, a more conservative testis-sparing surgery was proposed, but it requires a preliminary differentiation between benign and malignant masses: this however remains challenging. Although common understanding in radiology and oncology is that perfusion patterns might provide a useful information about the type of masses, no guidelines or consensus is available for the differentiation of STMs. We propose to build a dictionary of relevant perfusion patterns, extracted using non-negative matrix factorization on pixel-wise time-intensity curves from contrast-enhanced ultrasound data. When data from a lesion are reconstructed using this dictionary, a vector containing the frequency of utilization of each pattern can be used as a tissue signature. Using this signature, a support vector machine classifier has been trained, and the cross validated accuracy reached 100% in our pilot cohort.

KW - Cancer

KW - CEUS

KW - Dictionary learning

KW - SVM

KW - Ultrasound non-negative matrix factorization

UR - http://www.scopus.com/inward/record.url?scp=85073905299&partnerID=8YFLogxK

U2 - 10.1109/ISBI.2019.8759528

DO - 10.1109/ISBI.2019.8759528

M3 - Conference paper

AN - SCOPUS:85073905299

VL - 2019-April

SP - 850

EP - 854

BT - ISBI 2019 - 2019 IEEE International Symposium on Biomedical Imaging

PB - IEEE Computer Society

T2 - 16th IEEE International Symposium on Biomedical Imaging, ISBI 2019

Y2 - 8 April 2019 through 11 April 2019

ER -