Reinforcing Digital Pathology as the new Big Data
Microscope
Illustration with two recent studies
Nicolas Loménie
http://www.math-info.univ-paris5.fr/~lomn/
December 19
th, 2017
The Big Picture
Mainstream Research Line in Medical Imaging
2/15
Mohammad Havaei, Axel Davy, David Warde-Farley, Antoine Biard, Aaron Courville, Yoshua Bengio, Chris Pal, Pierre-Marc Jodoin, Hugo Larochelle, Brain tumor segmentation with Deep Neural Networks, Medical Image Analysis, Volume 35, January 2017, Pages 18-31
The Big Picture
But we cannot model histopathological images the same way
3/15
The Big Picture
But we cannot model histopathological images the same way
3/15
Outline
•
Building up a community ;
•
Defining new methodologies or Reinforcing existing ones ? ;
4/15
Building up a community
First Contact - 2009-2012
PUPH Frédérique Capron - Hospital La Pitié Salpêtrière
•
In Singapore, UMI IPAL -> ANR MICO (2010-2013) project with Frédérique Capron and
•
Explore the concept of cognitive microscopy
5/15
Building up a community
Whole Slide Image (WSI)
Biopsy glass slide
Ü Scan at
40X Resolution :
0.2456µm per pixel
Ü
Whole Slide Image 8 GPixels
•
Multi-page TIFF image
•
File size (compressed) : Usually 1 to 2 GB
•
Image size (uncompressed) : 15 to 25 GB
5/15
Mitosis Detection Contest
An ICPR 2012 Contest
5/15
Roux Ludovic, Racoceanu Daniel, Loménie Nicolas, Kulikova Maria, Irshad Humayun, Klossa Jacques, Capron Frédérique, Genestie Catherine, Le Naour Gilles, N Gurcan Metin Mitosis detection in breast cancer histological images : An ICPR 2012 contest J Pathol Inform 2013, 4 :8 (30 May 2013)
Challenges (An ICPR 2012 Contest)
125 registered contestants in 39 countries - IPAL lab - Singapore
5/15
Cited in IEEE Journal of Biomedical and Health Informatics, vol. 18, no. 2, March 2014 : Toward Automatic Mitotic Cell Detection and Segmentation in Multispectral Histopathological Images.
67 citations up to now
Mitosis Detection + Nuclear Atypia Grand Challenge etc.
5/15
Building up a community
2017 Refueling it : Master Internship : Denfeng CAO - Master BME (2017) HEGP - INSERM / PUPH Cecile Badoual
•
Objective : Automated HPV Cancer analysis + Spatial analysis of cell microenvironement
•
Correlate immune response and micro-environment analysis of the cell to cancer pronostic
5/15
Building up a community
INFORM software
•
Objective : Automated HPV Cancer analysis + Spatial analysis of cell microenvironement
5/15
Unité de Technologies Chimiques et Biologiques pour la Santé - CNRS /INSERM
•
PUPH Jean-François Emile (Ambroise Paré) - DR Nathalie Mignet
•
Objective : Lymphocyte Infiltration : VISILOG software
6/15
UTCBS - CNRS /INSERM
Master Internship : Tiede Armand DJIRO - Master Plurimedia (2017)
•
PUPH Jean-François Emile (Ambroise Paré) - DR Nathalie Mignet
•
Objective : Lymphocyte Infiltration : VISILOG software
•
Analysis of the micro-environment of the tumor (mice model)
6/15
CEA - MIRCen Thierry Delzescaux
PhD Mid-term committee - Clément Bouvier
7/15
Science. 2013 Jun 21 ;340(6139) :1472-5. doi : 10.1126/science.1235381.
BigBrain : an ultrahigh-resolution 3D human brain model. Amunts K1, Lepage C, Borgeat L, Mohlberg H, Dickscheid T, Rousseau MÉ, Bludau S, Bazin PL, Lewis LB, Oros-Peusquens AM, Shah NJ, Lippert T, Zilles K, Evans AC
Sainte-Anne - INSERM
PUPH Catherine Oppenheim and PhD. MD. Pascale Varlet
•
Objective : IMABRAIN (Associate Member)
•
Multi-scale analysis of the brain (pathology and development)
7/15
Outline
•
Building up a community ;
•
Defining new methodologies or Reinforcing existing ones ? ;
8/15
Methodological innovation...
Graph-based analysis and Digital Histopathology ?
9/15
A Hybrid Classification Model for Digital Pathology Using Structural and Statistical - Pattern Recognition - Erdem Ozdemir and Cigdem Gunduz-Demir -
IEEE Transactions on Image Processing - vol. 32, no. 2, February 2013
...but we need basic,efficient software first...
10/15
... leading to immediate valorisation
Figure:Diagram flow of the entire process line from WSI processing to infiltration curve generation
11/15
... leading to immediate valorisation
with open solution like Cytomine
Figure:Annotation of the tumor boundary
11/15
... leading to immediate valorisation
Figure:Illustration of lymphocyte detection. (a) RGB image of colorectal carcinoma. (b) staining séparation. (c) Otsu and K-means (two clusters)
11/15
... leading to immediate valorisation
-> Innovation : SATT, Companies, Companion Test (Nov. 2017)
Figure:Illustration of creating similar borders. (a) Annotation of a tumor boundary identical to the ROI. (b) Normalised outside distance card. (c) Normalised inside distance card. (d) and (e) similar borders by distance 5µm (configurable).
11/15
... and more to come,
possibly a portfolio by institutional partner.
Figure:Illustration of the process of creating curves. (a) Annotation of tumor border and region of interest (ROI), (b) treatment of ROI and creation of similar border at tumor border, (c) detection of lymphocytes, (d) repositioning of ROI and assignment of pixels and (e) the
quantitation curve of lymphocyte infiltration.
12/15
... and more to come,
possibly a portfolio by institutional partner.
12/15
... and more to come,
possibly a portfolio by institutional partner.
12/15
... and more to come,
possibly a portfolio by institutional partner.
12/15
Wrap up
•
Emerging field ;
•
More and more graph-based analysis ;
•
Emerging Community : http://www.math-info.
univ-paris5.fr/~lomn/Workshop/WorkshopHisto.html
13/15
Lomenie and Racoceanu. (2013).Point set morphological filtering and semantic spatial configuration modeling: application to microscopic image and bio-structure analysis,Pattern Recognition,45(8) :2894-2911.
Ta, Lezoray, Elmoataz and Schupp. (2009).Graph-based Toolsfor Microscopic Cellular Image Segmentation,Pattern Recognition,42(6) :1113-25.
Wrap up
•
Emerging field ;
•
More and more graph-based analysis ;
•
Emerging Community : http://www.math-info.
univ-paris5.fr/~lomn/Workshop/WorkshopHisto.html
13/15
Andrew Janowczyk, Scott Doyle, Hannah Gilmore, Anant Madabhushi -A resolution adaptive deep hierarchical (radhical)learning scheme applied to nuclear segmentation of digital pathology images - 2016/4/1 - Computer Methods in Biomechanics and Biomedical Engineering : Imaging &
Visualization
Basavanhally, Ganesan, Agner, Monaco, Feldman, Tomaszewski. (2010) Computerized image-based detection andgrading of lymphocytic infiltrationin HER2+ breast cancer histopathology,IEEE Trans Biomed Eng.,57:642-653.
Wrap up
•
Digital histopathology and deep learning coming up ;
•
Databases under construction with challenges :
• http://ccdb.ucsd.edu/
• https://cancergenome.nih.gov
• https://tma.im/cgi-bin/home.pl
• http://web.inf.ufpr.br/vri/breast-cancer-database
13/15
When machine vision meets histology: A comparative evaluation of model architecture for classification of histology sections - Zhonget al.- Medical Image Analysis - 2017, Volume 35, Issue 3, Pages 530–543
Horizon
14/15
Coming into Focus : Computational Pathology asthe New Big Data Microscope- The American Journal of Pathology - Cellular and Molecular Biology of disease - Kevin A. Roth, r Kevin A. Roth - March 2015, Volume
185, Issue 3, Pages 600–601
Thank You for your Attention
We miss grant call for us
Nicolas Loménie, Eng., Ph.D., HDR
Laboratoire d’Informatique de Paris Descartes (LIPADE)
Equipe d’Acceuil EA 2917
Groupe Systèmes Intelligents de Perception (SIP) http ://w3.mi.parisdescartes.fr/sip-lab/
&
IMABRAIN - INSERM
Hospital Sainte Anne
15/15