About This Project
Breast cancer is the second most common cause of cancer death among women, and requires careful histologic study for diagnosis. I am studying the feasibility of recognizing breast cancer cells in histologic specimen by means of artificial intelligence. We will train convolutional neural networks to recognize tumour cells in small snippets of large images. If positive, the region in the image corresponding to the snippet is marked. Our preliminary results are promising (see also lab notes)!
Ask the Scientists
Join The DiscussionWhat is the context of this research?
Breast cancer is the most common cancer type among women and the second most common cause of cancer death among women. Diagnosis of breast cancer is established using several different methods: physical examination, radiology and histology.
During histologic study, breast tissue is biopsied and examined by a pathologist. This is a critical step in correctly diagnosing cancerous cells and planning necessary treatment.
The human eye is able to recognize a large variety of pathologies, but the procedure can be time consuming and error-prone. Careful examination of histological material is critical to correct diagnosis and treatment of patients.
What is the significance of this project?
During the past decade there has been a rapid evolution of convolutional neural networks (https://en.wikipedia.org/wiki/...). This form of artificial intelligence has been applied to medical imaging and histopathology to identify mass lesions in mammography [5], with promising results.
Automated detection of tumour cells in histologic specimen will help pathologists scan more samples in less time and focus on difficult cases. In some cases thorough histological examinations can take up to one hour.
This technology could decrease the cost associated with breast cancer screening and allow for more thorough screening of biopsied tissue.
Moreover, this project should also provide us with the necessary knowledge to apply our findings to other types of tumours.
What are the goals of the project?
The goal of our project is to provide a faster mean of detection of cancer cells in histologic specimen. We believe that AI can be used to identify cancer cells in histologic specimens with as much accuracy as the human eye.
We will test this by training convolutional neural networks to identify breast cancer cells in images of prepared samples, and comparing the success rate of our CNN to that of humans identifying cancerous tissue. We have created an image database of human breast cancer and normal breast tissue. We have created the necessary programs for the training of the convolutional neural networks (CNNs) and we have set up a database for the results.
Budget
Training of Neural Network is done by complex algorithms. They use millions of variables to train the neural network. The process of training is computationally demanding. Although graphic cards are often used for gaming, their internal layout is a very powerful parallel processor with many thousands of compute cores that can perform the necessary calculations faster. Without a graphics card, the necessary calculations might take years instead of weeks. As we are analyzing medical data, we are not permitted to transfer the images to other computers where such graphic cards are already installed. No manpower costs will be covered by this item.
Endorsed by
Project Timeline
Research has begun as of Feb 2017. Until now the necessary computer programs have been developed and an image database for histological specimen has been created (after approval of the competent ethical review committee).
Until summer 2018 the training of the neural networks should be accomplished.
Results should be published in Autumn 2018.
Mar 01, 2018
Program development
Apr 09, 2018
Project Launched
Sep 01, 2018
Finish training
Nov 01, 2018
Dissemination/Publication
Meet the Team
Affiliates
Team Bio
Team:
Our team consists of a pathologist (Dr. Milo Halabi, Hospital Ried/Innkreis, Austria), a professor for informatics (Prof. Wilfried Gansterer, University of Vienna), a specialist for the application of neural networks in medicine (Dr. Roxane Licandro, Technical University of Vienna) and me.
Robert Fritze
Motivation:
I believe Informatics has the potential to make valuable contributions towards my aim of developing medicine to alleviate the patient afflictions. Therefore once finished my studies in medicine I have started to study informatics. During the study program I became aquainted with many new methods and algorithms. My experience in the medical field helps me to identify current problems and find solution strategies. Many of my research ideas are too complex to be worked on by a single person. In this case I try to compile a team of specialists.
Education:
2002 European primary medical qualification, University of Vienna
2012 European specialist medical qualification (anaesthesiology and intensive care medicine)
This project is part of the master’s thesis I am currently writing (in the field of Scientific Computing) at the University of Vienna and I expect to finish the program by Autumn 2018.
Publications:
Robert Fritze, Anita Graser, Markus Sinnl, Combining spatial information and optimization for locating emergency medical service stations: A case study for Lower Austria, International Journal of Medical Informatics, Volume 111, March 2018, Pages 24-36, see http://www.loweraustriaemsresu...
Additional Information
Preliminary reseach:
[1] Litjens, G. et al. Deep learning as a tool for increased accuracy and efficiency of histopathological diagnosis. Sci. Rep. 6, 26286;
[2] Wang H et al. Mitosis detection in breast cancer pathology images by combining handcrafted and convolutional neural network features. Journal of Medical Imaging. 2014;1(3)
For a general introduction to convolutional neural networks and artificial intelligence see:
[3] Goodfellow et al., Deep Learning, MIT Press 2016, ISBN 9780262035613
For an introduction to machine learning (of which artificial intelligence is a part) see:
[4] Bishop, Pattern recognition and machine learning, Springer 2006, ISBN-13: 978-0-387-31073-2
Further literature:
[5] J. Arevalo, F. A. González, R. Ramos-Pollán, J. L. Oliveira and M. A. Guevara Lopez, "Convolutional neural networks for mammography mass lesion classification," 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Milan, 2015, pp. 797-800.doi: 10.1109/EMBC.2015.7318482
The cover image was downloaded from Wikipedia on March 24th 2018, Licensing notice: Difu Wu, S10-5263 H&E 20x DCIS, CC BY-SA 3.0
Project Backers
- 4Backers
- 14%Funded
- $115Total Donations
- $28.75Average Donation