Skip to Main content Skip to Navigation
New interface

Contribution au mosaïquage d'images endoscopiques d'organes creux pour l'aide au diagnostic du cancer en phase précoce

Abstract : This PhD thesis is a contribution to the development of a method of cartography of the hollow organs internal walls based on video-endoscopic image sequences, applicable to cancerous lesion localization inside these cavities (bladder in particular). This approach should enable the specialists to better choose the biopsy sites and therefore enhance the detection sensibility. Such constructed maps should facilitate post-operative controls and reports. The first part of this PhD thesis deals with the physical and instrumental principles allowing the detection of precancerous lesions using multiple bands autofluorescence imaging. This technique lies on the fact that the autofluorescence spectrum emitted by a healthy tissue under appropriate excitation is different from that emitted by a cancerous tissue. For this technique, two images of the same region to be examined are acquired. An image, resulting from the pixel wise division of the intensities of the two acquired images, highlights the suspicious lesions. A specific graphical interface was developed allowing us to control the optical-mechanical system of acquisition of autofluorescence images at two different wavelengths and to visualize the images in various operating modes: white light, autofluorescence and/or compound (intensity ratio images).The second part of this PhD thesis presents the method developed for building a map of the examined region using image sequences. We have implemented a projective transformation model. A grey-level image registration algorithm, based on mutual information and on a stochastic steepest gradient descent technique, was implemented to estimate perspective transformation parameters linking geometrically each pair of consecutive images of the sequence. These transformations authorize a map building using a sequential mosaicing method. Correction methods of the defaults of endoscopic images (radial distortion, endoscope's fibre pattern, etc.) and the diminution of the registration errors accumulated through the perspective transformations are also presented. The results obtained on test images of pig bladder enabled to validate the robustness of our automatic registration procedure but with still long processing times. The precision obtained after cumulated error correction is of a few pixels. The registration method developed also allowed us to propose a simplified procedure for endoscope calibration.
Document type :
Complete list of metadata
Contributor : Thèses UL Connect in order to contact the contributor
Submitted on : Thursday, March 29, 2018 - 12:35:34 PM
Last modification on : Saturday, October 16, 2021 - 11:14:14 AM

Intranet access


  • HAL Id : tel-01749998, version 1



Rosebet Miranda Luna. Contribution au mosaïquage d'images endoscopiques d'organes creux pour l'aide au diagnostic du cancer en phase précoce. Autre. Institut National Polytechnique de Lorraine, 2005. Français. ⟨NNT : 2005INPL025N⟩. ⟨tel-01749998⟩



Record views