Abstract 4292: Fast and automated assessment of tumor response: Infrared imaging

Abstract

Abstract The rabbit Vx2 tumor is a fast-growing carcinoma model, which is commonly used to study different aspects of tumor behaviour under cancer treatments. The reduction of tumor viability and the degree of induced necrosis are the most common criteria to evaluate the efficacy of cancer treatments. The most recent developments in infrared microspectroscopy (IRMS) imaging aimed at automating the procedure of tissue recognition and quantification by using statistical methods and prediction algorithm. We used IRMS for the automatic characterization and quantification of the Vx2 liver tumor viability after a chemoembolization treatment. Twenty-eight rabbits with Vx2 liver tumor were included in this study: 20 rabbits were subjected to a Doxorubicin eluting beads (DEB) treatment and compared to a control (CTRL) group of 8 rabbits. The tumor bearing livers were resected, fixed in formalin and embedded in paraffin. Two adjacent sections were cut from each sample using a microtome. The first section was mounted on a calcium fluoride window suitable for IRMS imaging. The second section was put on a standard glass slide and stained with HES to serve as a control for IRMS imaging. On a first series of 14 different tumor sections (CTRL: 7 sections, DEB: 7 sections), we developed and validated a prediction algorithm. The protocol consisted of K-means (KM) clustering followed by linear discriminant analysis (LDA). The KM clustering was used to classify the spectra from the infrared images and to build a data base containing a large number of reference spectra (397.289 spectra) characteristics for tumor necrosis, viable tumor, fibrosis, liver parenchyma and liver parenchyma necrosis. Once the model validated, we empirically attributed a color for each type of tissue. Then, the predictive model was applied to infrared images of 52 new test tumor sections (CTRL: 9 sections, DEB: 43 sections). The result of the LDA model analysis is a false color image where each color corresponds to a type of tissue. The percentage of pixels corresponding to each color is automatically recorded by the LDA model. This advantage was used to calculate the mean percentage of surface occupied by each type of tissue and to determine the tumor viability after the DEB treatment. The LDA false color images reproduced the histological structures of Vx2 liver tumors. The sensitivity and specificity of the LDA model were high to 86.7% and 96.7% for the 5 tissue types respectively. For DEB group, the LDA model determined that the surface of necrotic tissue represented 77.68±23 % (CTRL group: 16.89±9 %, Mann Whitney: P<0.0001), the viable tumor 14.29±23 % (CTRL group: 74.74±7 %, MW: P< 0.0001) and fibrosis 3.89±6 % of the tumor (CTRL group: 1.1±2 %, MW: P= 0.6262). The remaining percentage corresponded to unclassified spectra (DEB group: 3.94±7 %, CTRL group: 6.40±5 %, MW: P=0.0135). Our results show that IR imaging coupled with LDA model analysis could be a helpful to easily assess tumor response. Citation Format: Hadrien D'inca, Florentina Pascale, Saida Homayra Ghegediban, Michel Wassef, Cyril Gobinet, Julien Namur, Alexandre Laurent, Michel Manfait. Fast and automated assessment of tumor response: Infrared imaging. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 4292. doi:10.1158/1538-7445.AM2014-4292