Final Activity Report Summary - IRONDB (MR-based analysis, Indexing, and Retrieval of brain irON Deposition in Basal ganglia)
The IRONDB project has been executed as planned in both time and deliverables. We recruited three fellows (Experienced Researchers) in time aligned with our plans. For all positions, there were many applicants showing the appeal of the project and collaboration. For Fellow3, the eligibility requirements of being a non-Turkish person and having at least 2 years of industrial experience reduced our candidate pool considerably. In the end, for that position as well, we could hire a very strong candidate. In the recruitment stage, we received feedback from both partners and sought the approvals of both Prof. Aytul Ercil and Prof. Mujdat Cetin. During the execution, we decided to cancel the shorter term stays and informed this in the previous reports. The cancellation has not affected the overall progress because we took measures to compensate for this by making short trips as well as having weekly teleconferences.
For the scientific part, we have fulfilled the deliverables for the seven main workpackages we had described in the project proposal. These tasks include:
1. Collection of the dataset and ground truth,
2. Normalisation of the MR data and detection of hypo-intense voxels.
3. Organ segmentation.
4. Shape-based search and retrieval.
5. Expert-knowledge for semantic search and retrieval.
6. Visualisation.
7. Validation, demonstration, and documentation.
We used both proprietary and public datasets for the development and testing of our algorithms. We initiated new clinical contacts, such as Anatolian Health Center, Istanbul and Yeditepe University Hospital, Istanbul. We could formalise our collaboration with Yeditepe University Hospital by taking part in a Clinical Investigation Agreement between Philips Healthcare and the university. To further improve the collaboration, especially at the technical level, an ftp site has been set up to share data, software, reports, and the experimental results. The fellows learned a great deal from the interactions with clinical experts and also made publications with them. We considered how to deploy our system in a hospital setting. We conducted a clinical validation of our search and retrieval system and showed that our system can improve the diagnostic decision process of radiologists.
Throughout the project, we emphasised the collaboration among fellows despite the geographical distance between the partners. With visits, teleconferences, data and knowledge sharing, and common project definitions, we created ways to enhance this. For example, in early 2009, we started a collaborative activity on the GUI design among the three fellows and Dr Ekin. Teleconferences were held, a source repository was established at Philips, and new GUI was designed. In the transfer of Knowledge phase, Dr. Devrim Unay shared his knowledge in search and retrieval schemes with his colleagues at Sabanci University. This resulted in a joint paper published in the ImageCLEF contest, where the joint team received the fourth place. In total, 12 of 19 conference papers (more than 60 percent) included more than one fellow in the author list quantitatively proving the collaborations in the project.
We were active in making publications in leading journals and conferences. Four journal papers were direct result of the fellows' work during their recruitment period; several more resulted from the initiatives the fellows took beforehand and continued during their tenures. The number of peer-reviewed publications was 23 (four journal, 19 conference papers) in 60 man-months making one publication per less than three man-months of the project. We also received interest from Philips Intellectual Property Office that filed three patent applications for our work. Overall, the project has achieved her aims by increasing the depth of brain MR analysis knowledge in both partners and by providing a common framework for them to share their knowledge that had been built before.
For the scientific part, we have fulfilled the deliverables for the seven main workpackages we had described in the project proposal. These tasks include:
1. Collection of the dataset and ground truth,
2. Normalisation of the MR data and detection of hypo-intense voxels.
3. Organ segmentation.
4. Shape-based search and retrieval.
5. Expert-knowledge for semantic search and retrieval.
6. Visualisation.
7. Validation, demonstration, and documentation.
We used both proprietary and public datasets for the development and testing of our algorithms. We initiated new clinical contacts, such as Anatolian Health Center, Istanbul and Yeditepe University Hospital, Istanbul. We could formalise our collaboration with Yeditepe University Hospital by taking part in a Clinical Investigation Agreement between Philips Healthcare and the university. To further improve the collaboration, especially at the technical level, an ftp site has been set up to share data, software, reports, and the experimental results. The fellows learned a great deal from the interactions with clinical experts and also made publications with them. We considered how to deploy our system in a hospital setting. We conducted a clinical validation of our search and retrieval system and showed that our system can improve the diagnostic decision process of radiologists.
Throughout the project, we emphasised the collaboration among fellows despite the geographical distance between the partners. With visits, teleconferences, data and knowledge sharing, and common project definitions, we created ways to enhance this. For example, in early 2009, we started a collaborative activity on the GUI design among the three fellows and Dr Ekin. Teleconferences were held, a source repository was established at Philips, and new GUI was designed. In the transfer of Knowledge phase, Dr. Devrim Unay shared his knowledge in search and retrieval schemes with his colleagues at Sabanci University. This resulted in a joint paper published in the ImageCLEF contest, where the joint team received the fourth place. In total, 12 of 19 conference papers (more than 60 percent) included more than one fellow in the author list quantitatively proving the collaborations in the project.
We were active in making publications in leading journals and conferences. Four journal papers were direct result of the fellows' work during their recruitment period; several more resulted from the initiatives the fellows took beforehand and continued during their tenures. The number of peer-reviewed publications was 23 (four journal, 19 conference papers) in 60 man-months making one publication per less than three man-months of the project. We also received interest from Philips Intellectual Property Office that filed three patent applications for our work. Overall, the project has achieved her aims by increasing the depth of brain MR analysis knowledge in both partners and by providing a common framework for them to share their knowledge that had been built before.