Spectral features selection and classification for bimodal optical spectroscopy applied to bladder cancer in vivo diagnosis

Emilie Pery (1) , Walter Blondel (1) , Samy Tindel (2) , Maha Ghribi (2) , Agnès Leroux (3) , François Guillemin (3, 1)

1 CRAN - Centre de Recherche en Automatique de Nancy
2 IECL - Institut Élie Cartan de Lorraine
3 CAV - Centre Alexis Vautrin

This paper describes an experimental study combining spatially resolved AutoFluorescence (AF) and Diffuse Reflectance (DR) fibred spectroscopies to discriminate in vivo between healthy and pathological tissues in a preclinical model of bladder cancer. Then, a detailed step-by-step analysis scheme is presented for the extraction and the selection of discriminative spectral features (correlation, Linear Discriminant and Logistic Regression Analysis), and for the spectroscopic data final classification algorithms (Regularized Discriminant Analysis and Support Vector Machines). Significant differences between healthy, inflammatory and tumoral tissues were obtained by selecting a reasonable number of discriminant spectral features from AF, DR and Intrinsic Fluorescence spectra, leading to improved sensitivity (87%) and specificity (77%) compared to monomodality (AF or DR alone).

Mots clés: Absorption; Bladder; Cancer; Feature extraction; In vivo; Spectroscopy; supervised classification; bladder cancer; autofluorescence; diffuse reflectance; Tumors; feature selection

Domaines: Ingénierie biomédicale

Emilie Pery, Walter Blondel, Samy Tindel, Maha Ghribi, Agnès Leroux, et al.. Spectral features selection and classification for bimodal optical spectroscopy applied to bladder cancer in vivo diagnosis. IEEE Transactions on Biomedical Engineering, 2014, 61 (1), pp.207-216. ⟨10.1109/TBME.2010.2103559⟩. ⟨hal-00541125⟩