Section: New Results

Biological Image Analysis

Content-based Image Retrieval for the Early Diagnosis of Gastrointestinal Cancer based on Optical Biopsies

Participants : Barbara André [ Correspondant ] , Tom Vercauteren [ Mauna Kea Technologies ] , Nicholas Ayache.

In vivo pathology from endomicroscopy videos can be a challenge for many physicians. To ease this task, we propose a content-based video retrieval method providing, given a query video, relevant similar videos from an expert-annotated database (see Fig. 9 ). Our main contribution consists in revisiting the Bag of Visual Words method by weighting the contributions of the dense local regions according to the registration results of mosaicing. We perform a leave-one-patient-out k-nearest neighbors classification and show a significantly better accuracy (e.g. around 94% for 9 neighbors) when compared to using the video images independently. Less neighbors are needed to classify the queries and our signature summation technique reduces retrieval runtime [46] , [47] .

Figure 9. Typical mosaic retrieval results provided by our method from one benign query. B. indicates Benign and N. Neoplastic. From left to right on each row: the queried mosaic, and its k-NNs on the top layer, and their respective colored visual words on the bottom layer. FOV of the mosaics: from 260 microns to 1300 microns.

Pre-clinical molecular imaging

Participants : Marine Breuilly [ Correspondant ] , Grégoire Malandain, Nicholas Ayache, Jacques Darcourt [ CAL ] , Philippe Franken [ CAL ] , Thierry Pourcher [ CEA ] .

The aim of this project is to track and to quantify the cancer growth in mice with a coupled SPECT/CT imaging device. The collaboration with the Transporter in Imagery and Oncologic Radiotherapy team (TIRO, UNSA) allow a hand-in-hand work, from the planning of the experiment and manipulation of the animal to the study of the evolution of the disease through the acquisition of SPECT and CT images. Among others, the respiratory motion will be studied as metastasis can appear in organs surrounding the diaphragm (see Fi. 10 ), since this motion can lead to blurred acquisitions.

Figure 10. Tranverse, sagittal and coronal slices of a SCID mouse depicting the fusion of a CT image with SPECT data acquired with GE eXplore speCZT CT 120: hot spots reveal liver metastasis from adenocarcinoma of the colon.

Microscopy image reconstruction and automatic lineage tracking of the growing meristem cells

Participants : Romain Fernandez [ Correspondant ] , Grégoire Malandain, Christophe Godin [ EPI Virtuals Plants ] , Jean-Luc Verdeil [ Cirad ] , Jan Traas [ INRA/ENS Lyon ] , Pradeep Das [ ENS Lyon ] .

We studied the tracking of meristem cells using time-lapse confocal microscopy acquisition on early stages flowers of Arabidopsis shoot apical meristems. We designed a reconstruction method (MARS) and a tracking algorithm (ALT) in order to map the segmentations of the same meristem at different times, based on a network flow representation in order to solve the cell assignment problem (see results in Fig. 11 ). We validated the MARS-ALT pipeline on a four-steps timecourse of an early stage floral bud. The validation by biologists showed the efficiency of the segmentation algorithm on the reconstructed images (near to 96% of well-identified cells) and of the lineaging algorithm (100% of well-identified lineages in the easiest case and 90% in the hardest).

Figure 11. The Automated Lineage Tracking algorithm identifies succesfully most of the descendances (image on the right) of the mother cells (image on the left) of a full Arabidopsis floral bud, despite the 24 hours between the two acquisitions and the big amount of deformations and divisions. The colormap is randomly generated in order to validate visually the detected lineages by comparing the colors on both images.