On feature extraction and classification in prostate cancer radiotherapy using tensor decompositions

• Published in: Medical engineering & physics Vol. 37; no. 1; pp. 126 - 131
• Main Authors: Fargeas, Auréline, Albera, Laurent, Kachenoura, Amar, Dréan, Gaël, Ospina, Juan-David, Coloigner, Julie, Lafond, Caroline, Delobel, Jean-Bernard, De Crevoisier, Renaud, Acosta, Oscar
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01-01-2015; Elsevier
• Subjects: Algorithms; Bioengineering; Bleeding; Cancer; Canonical polyadic decomposition; Classification; Deterministic multi-way analysis; DIAG; Diagnosis, Computer-Assisted - methods; Discriminant Analysis; Dose-Response Relationship, Radiation; Gastrointestinal Hemorrhage - diagnosis; Gastrointestinal Hemorrhage - etiology; Human health and pathology; Humans; Life Sciences; Linear Models; Male; Mathematical analysis; Patients; Prediction of toxicity; Principal Component Analysis; Probability; Prognosis; Prostate; Prostate cancer; Prostatic Neoplasms - radiotherapy; Radiation Injuries - diagnosis; Radiation Injuries - etiology; Radiology; Radiotherapy; Radiotherapy Dosage; Radiotherapy Planning, Computer-Assisted; Rectum; Risk; Sensitivity and Specificity; Support Vector Machine; Toxicity; Urology and Nephrology; Canonical polyadic decomposition; Prediction of toxicity; Classification; DIAG; Radiotherapy; Prostate cancer; Deterministic multi-way analysis
• ISSN: 1350-4533; 1873-4030
• DOI: 10.1016/j.medengphy.2014.08.009

Highlights • Classification of patients at risk of presenting rectal bleeding after an external beam radiotherapy, which is commonly prescribed for prostate cancer. • Presentation of a novel method of classification based on a Deterministic Multi-way Analysis (DMA) of three-dimensional planned dose distributions across a population. • Computation of two bases of vectors representing bleeding and non bleeding patients by using the Canonical Polyadic (CP) decomposition of two fourth order arrays of the planned doses. • Classification of a patient according to its distance to the subspaces spanned by both bases. • Good specificity and sensitivity values of the proposed CP-DMA method, outperforming the classical approaches.