6533b7d5fe1ef96bd1264d3e

RESEARCH PRODUCT

Color and multispectral image processing for the detection of inflammatory lesions of the stomach

subject

description

The work presented in this manuscript is part of the ANR project EMMIE. This project aims to develop an innovative multimodal system for the detection of inflammatory lesions in the stomach. To this purpose, a prototype has been developed to be able to acquire NBI endoscopic images and multispectral images during human's antrum exploration. The prototype is made of a standard endoscope and multispectral images.The prototype can acquire two types of data: NBI images and spectra. These two modalities are processed independently. Common image processing features are used to recognize four kind of diseases: active gastritis, chronic gastritis, metaplasia and atrophy. In addition, visual based features have been developed with an experienced endoscopist. These criteria were successfully used to classify endoscopic image patches.NBI images are pre-processed with "intrinsic image decomposition". This decomposition is made by the dichromatic model. We developed two methods to inverse this model, respectively with quadratic programming and deconvolutions. Our algorithms are tested on three datasets. Equivalent or better results are found compare to the state of the art.Acquisitions are also made on mice model. Reflectance spectra have been acquired on resected stomach from control and inflammed mice. A pipeline was build to classify the two groups. The pipeline allowed us to obtain good classification results and to identify two interesting wavelengths range in the near infrared.This pipeline is re-used to process~ extit{in vivo} spectra from human's stomach with the developed acquisition prototype. The aim is to characterize the inflammation and being able to classify spectra according to the disease. Obtained results are compared with biopsies and show good diagnostic prognosis for chronic gastritis and intestinal metaplasia.