PhD Thesis
Automated Mitosis Detection in Color and Multi-spectral High-Content Images in Histopathology: Application to Breast Cancer Grading in Digital Pathology
This work has been done in International French-Singaporean lab (IPAL) in Singapore, with the support of prestigious partners[1] (NUS, A*STAR, CNRS, UPMC, IMT and UJF). The work has been done in the framework of the ANR (The French National Research Agency – Technologies for health and autonomy – TechSan) project entitled MICO[2] (COgnitive virtual MIcroscopy for Breast Cancer Grading), with reputable collaborators as THALES, AGFA healthcare, TRIBVN and the Hospital “Pitié-Salpêtrière” from Paris, the most important European hospital. The PhD degree was delivered by the MSTII doctoral school (Mathematics, Science and Technology of Information, Informatics) of the UJF from Grenoble, France.
Abstract:
Digital pathology represents one of the major and challenging evolution in modern medicine. Pathological exams constitute not only the gold standard in most of medical protocols, but also play a critical and legal role in the diagnosis process. Diagnosing a disease after manually analyzing numerous biopsy slides represents a labor-intensive work for pathologists. Thanks to the recent advances in digital histopathology, the recognition of histological tissue patterns in a high-content Whole Slide Image (WSI) has the potential to provide valuable assistance to the pathologist in his daily practice. Histopathological classification and grading of biopsy samples provide valuable prognostic information that could be used for diagnosis and treatment support. Nottingham grading system is the standard for breast cancer grading. It combines three criteria, namely tubule formation (also referenced as glandular architecture), nuclear atypia and mitosis count. Manual detection and counting of mitosis is tedious and subject to considerable inter- and intra-reader variations. The main goal of this dissertation is the development of a framework able to provide detection of mitosis on different types of scanners and multispectral microscope.
The main contributions of this work are eight fold. First, we present a comprehensive review on state-of-the-art methodologies in nuclei detection, segmentation and classification restricted to two widely available types of image modalities: H&E (Hematoxylin Eosin) and IHC (Immunohistochemical). Second, we analyse the statistical and morphological information concerning mitotic cells on different color channels of various color models that improve the mitosis detection in color datasets (Aperio and Hamamatsu scanners). Third, we study oversampling methods to increase the number of instances of the minority class (mitosis) by interpolating between several minority class examples that lie together, which make classification more robust. Fourth, we propose three different methods for spectral bands selection including relative spectral absorption of different tissue components, spectral absorption of H&E stains and mRMR (minimum Redundancy Maximum Relevance) technique. Fifth, we compute multispectral spatial features containing pixel, texture and morphological information on selected spectral bands, which leverage discriminant information for mitosis classification on multispectral dataset. Sixth, we perform a comprehensive study on region and patch based features for mitosis classification. Seven, we perform an extensive investigation of classifiers and inference of the best one for mitosis classification. Eight, we propose an efficient and generic strategy to explore large images like WSI by combining computational geometry tools with a local signal measure of relevance in a dynamic sampling framework. The real time evaluation of these frameworks is done in MICO (COgnitive MIcroscopy, ANR TecSan project) platform prototyping initiative.
We evaluated our proposed frameworks on MITOS ICPR 2012 contest[3] dataset. Compared to MITOS contest results, our multispectral framework outperforms significantly the best results, according to F-Measure. For the color framework; we manage to rank in second position with a F-Score very close to the first one, opening interesting hopes for the future of the cancer prognosis. Finally, our both frameworks manage to reach the same level of accuracy in mitosis detection on color and multispectral datasets, a promising result on the way on clinical routine use.
I actively involved in ITK[4] during my PhD. I proposed a novel operational framework related to Triangular / Simplex Mesh and Delaunay / Voronoi to scientific community working in medical imaging. I proposed an extension ofitk::QuadEdgeMesh data structure to handle both primal and dual meshes simultaneously. The new data structure, itk::QuadEdgeMeshWithDual, already includes by default the due topology, to handle dual geometry as well.
Figure 1. Automated Mitosis Detection Framework for Color Histopathology Images
Figure 2. Automated Mitosis Detection Framework for Multi-spectral Histopathology Images
Figure 3. Stereology flow used for mitosis score over a ROI in MICO platform.
[1] IPAL partners: http://ipal.cnrs.fr/our-supporting-institutes
[2] MICO project: http://ipal.cnrs.fr/project/mico-cognitive-virtual-microscopy
[3] ICPR 2012 MITOS Contest: http://ipal.cnrs.fr/ICPR2012/
[4] ITK (Insight Segmentation and Registration Toolkit): http://www.itk.org/