- Main research unit: Instituto de Telecomunicações (IT)
- Mário Jorge Dinis Ribeiro
- Paulo Jorge Afonso Alves
- Farhan Riaz
- Nuno Eduardo Roxo Rodrigues Cravo Barata
- Start date: 01.02.2011 | End date: 31.01.2014
- Financing: € 145,000.00 (FCT)
Project description: The main goal of the CAGE project is to explore Computer Vision and Human Computer Interaction methodologies in order to improve a clinician’s ability to diagnose Cancer quickly when facing a variety of Gastroenterology imaging modalities. While the goal of developing and deploying a full system for both clinical training and computer assisted decision is clearly too ambitious for a single project, our aims for this proposal are two-fold:
- Research computer vision algorithms that can extract clinical information from the visual data itself.
- Study adequate interaction mechanisms for integrating this type of technology into real gastroenterology environments.
We are motivated by the fact that cancer is a leading cause of death worldwide. From a total of 58 million deaths worldwide in 2005, cancer accounts for 7.6 million (or 13%) of all deaths [WHO08]. Gastric (or stomach) cancer is the second most lethal cancer in the World and that with highest mortality in the digestive tract among the Portuguese population. Every year approximately 1,000,000 new cases of gastric cancer are diagnosed worldwide [ACS08]. The follow-up of patients with lesions such as atrophy, intestinal metaplasia or dysplasia may lead into gastric cancer early diagnosis. If a cancer is detected early the prognosis is highly improved, motivating the development of systems that can support such detection or train clinicians to perform this task more efficiently.
Gastroenterology Imaging is today an essential tool for clinicians to detect cancer effectively. It is a rapidly evolving technological area with novel imaging devices such as Capsule, Narrowband Imaging or High-Definition Endoscopy. However, these technologies typically have a high time-price, even for an experienced clinician. The clear need for automation or semi-automation explains the increased relevance of the topic amongst Computer Vision scientists. In fact, doctors intuitively use visual features (colour and texture) in these procedures to diagnose several diseases. The medical community is aware that computer-assisted diagnosis will help and support a medical diagnosis preventing some errors and improving the health quality of their patients. Additionally, there are cases where clinical specialists are not available in the area, and therefore computer systems could provide a diagnosis to support the doctor's work. Finally, these systems could also teach new non-experienced medical doctors, in several areas of medicine.
State of the art in computer vision research already includes several high-quality publications for Gastroenterology such as topographic segmentation for capsule endoscopy [Coim06A] or event detection [ViIaO6]. However, some literature also shows that conventional research in more traditional scenarios such as multimedia archives does not translate well to in-body imaging scenarios given that these pose difficult challenges such as poor focus and illumination properties, reduced colour spaces, absence of geometrical structures, severe lens distortion, etc. [Coim06B]. In the CAGE project we will address issues such as rotation and scale invariant texture descriptors, illumination corrected colour descriptors, or visual and clinical data fusion for statistical pattern recognition.
On the other hand we are interested in understanding how these algorithms will translate into useful tools inside a Gastroenterology exam room. Which information shall we provide the clinician? How shall we display it? How will he interact with the system? Using vocal commands or gestures? The answers to these and other similar questions are essential for obtaining a clinical tool rather than a high-tech toy. Human Computer Interaction methodologies will be used during the CAGE project to ensure that we understand the best way that the user can access and benefit from this additional layer of information.
In order to accomplish these objectives we have assembled a multidisciplinary team of specialists, all of which have experience in research involving engineers, computer scientists and clinicians. This includes experts in Gastroenterology and medical technology validation (Dr. Mário Dinis Ribeiro, CINTESIS/FMUP/IPOPorto), Computer Vision and Human Computer Interaction (Dr. Miguel Coimbra, IT/FCUP, project's Principal Investigator - PI), and two PhD students (in Computer Vision).
The PI is a young Professor who has returned to Portugal after his PhD in London and has 9 years of experience in Computer Vision, 5 of which in Medical Imaging research. Dr. Miguel Coimbra leads a team of PhD and Msc students, has numerous publications on the field, leads one financed project and has participated as a researcher in 4 financed projects involving clinical partners. Finally, we have obtained the commitment of collaboration of a medical institution (IPOPorto), which will be an essential data provider and consultant.