Final Report Summary - MIRACLE (Microscopic Image Processing, Analysis, Classification and Modelling Environment)
The main objective of MIRACLE is to develop computer-aided diagnosis and analysis algorithms for microscopic image processing. To this day, analysis of histology images of the human tissue biopsies remains the most reliable way of diagnosing and grading cancer. Our work on the computer-aided classification and rating of histology images is motivated by the fact that there is significant inter- and intra-rater variability in the grading and diagnosis of cancer from histology slides by human experts. The goal is not to replace human experts with computers but develop algorithms and software to help pathologists and molecular biologists to produce more accurate grading and classification results. Therefore, computational histopathology can assist pathologists in making the grading and diagnosis reproducible and more accurate by providing useful quantitative measures from histology images of a patient suspected or diagnosed of having cancer.
In this exchange programme, young researchers were trained and educated in a virtual lab consisting of facilities from Bilkent University (BILKENT), University of Caen Basse-Normandie (UCBN), Information and Technology Institute - Centre for Research and Technology Hellas (CERTH-ITI), University of Warwick (WARWICK), and Ohio State University (OSU) in the US. As a result of this collaboration, microscopic image processing algorithms and software for images of Follicular Lymphoma (FL), neuroblastoma and cancer cell line images were developed, international workshops and special sessions in conferences were organized and joint scientific papers were published. We are in the process of organizing a special issue in Signal, Image and Video Processing (Springer) on the topic of microscopic image processing. This special issue will appear in late 2014 or early 2015.
In particular, we developed microscopic image representation and feature extraction schemes in WP1 based on colour content of FL images and cell-line images; microscopic image segmentation and model-based representation schemes in WP2; microscopic image classification methods in WP3; microscopic image data compression, storage and representation methods in WP4; and a microscopic image training database consisting of follicular lymphoma images and cancer cell-line images along with a semi-automatic annotation tool for FL images in WP5. One possible research avenue to develop algorithms and software is to mimic or follow the behaviour of pathologists and microbiologists to analyze the microscopic images. Another approach is to apply mathematical computer vision techniques such as texture classification and segmentation to analyze microscopic images, because the texture of cells is different in normal tissue and abnormal regions. One cannot favor one approach over the other, because computers do not have the intuition of people and people do not have the precise computation capability of computers. We believe that both approaches are valid and they have to be studied for each specific case.
The software, papers and theses can be downloaded from the web-page of MIRACLE. Cell-line images are also open to public. Researchers can upload their images to an on-line software and get their cell-line images classified. FL images can be inspected, annotated and downloaded by registering to the MIRACLE FL database maintained by ITI-CERTH. Additionally, medical experts can upload their own images expanding this way the available data.
Human cell shapes are different in different organs. As a result it is not possible to develop algorithms capable of dealing with all possible histopathological images. However, we believe that the systematic approach that we developed is applicable not only to FL images and some cancer cell-line images but also to other histopathological images.
Contact Information: Prof. A. Enis Cetin, Bilkent University, Electrical and Electronics Engineering Department, 06800 Ankara, Turkey
Telephone: +90-312-290-1477 Fax: +90-312-266-4192 E-mail: cetin[at]bilkent.edu.tr Website: and http://www.fp7-miracle.eu.
In this exchange programme, young researchers were trained and educated in a virtual lab consisting of facilities from Bilkent University (BILKENT), University of Caen Basse-Normandie (UCBN), Information and Technology Institute - Centre for Research and Technology Hellas (CERTH-ITI), University of Warwick (WARWICK), and Ohio State University (OSU) in the US. As a result of this collaboration, microscopic image processing algorithms and software for images of Follicular Lymphoma (FL), neuroblastoma and cancer cell line images were developed, international workshops and special sessions in conferences were organized and joint scientific papers were published. We are in the process of organizing a special issue in Signal, Image and Video Processing (Springer) on the topic of microscopic image processing. This special issue will appear in late 2014 or early 2015.
In particular, we developed microscopic image representation and feature extraction schemes in WP1 based on colour content of FL images and cell-line images; microscopic image segmentation and model-based representation schemes in WP2; microscopic image classification methods in WP3; microscopic image data compression, storage and representation methods in WP4; and a microscopic image training database consisting of follicular lymphoma images and cancer cell-line images along with a semi-automatic annotation tool for FL images in WP5. One possible research avenue to develop algorithms and software is to mimic or follow the behaviour of pathologists and microbiologists to analyze the microscopic images. Another approach is to apply mathematical computer vision techniques such as texture classification and segmentation to analyze microscopic images, because the texture of cells is different in normal tissue and abnormal regions. One cannot favor one approach over the other, because computers do not have the intuition of people and people do not have the precise computation capability of computers. We believe that both approaches are valid and they have to be studied for each specific case.
The software, papers and theses can be downloaded from the web-page of MIRACLE. Cell-line images are also open to public. Researchers can upload their images to an on-line software and get their cell-line images classified. FL images can be inspected, annotated and downloaded by registering to the MIRACLE FL database maintained by ITI-CERTH. Additionally, medical experts can upload their own images expanding this way the available data.
Human cell shapes are different in different organs. As a result it is not possible to develop algorithms capable of dealing with all possible histopathological images. However, we believe that the systematic approach that we developed is applicable not only to FL images and some cancer cell-line images but also to other histopathological images.
Contact Information: Prof. A. Enis Cetin, Bilkent University, Electrical and Electronics Engineering Department, 06800 Ankara, Turkey
Telephone: +90-312-290-1477 Fax: +90-312-266-4192 E-mail: cetin[at]bilkent.edu.tr Website: and http://www.fp7-miracle.eu.