Interactive Deep Learning for 3D Image Analysis
|Coordinator||Örebro universitet - Maskinteknik, Institutionen för naturvetenskap och teknik|
|Funding from Vinnova||SEK 500 000|
|Project duration||January 2017 - December 2017|
Purpose and goal
The purpose of the project was to develop an interactive learning-algorithm based on deep learning technology that will help medical experts to annotate CT data and involve them in the learning process of training phase algorithms. The goal was to obtain ethically approved data management, to introduce an intellectual property rights agreement between physicians and researchers, scientific dissemination, and to evaluate the user-friendliness and performance of a customized interactive annotation tool of CT data.
Expected results and effects
All objectives and expected effects for the project described in the proposal have been met. In addition to major scientific dissemination in the form of articles, workshops, conferences and newspapers, a newly established company involving researchers in computer science has been created for further development and application of the algorithms in medical and other applications. A collaboration between physicians and researchers has begun to investigate further issues beyond this project.
Planned approach and implementation
The project was conducted through physical meetings monthly with all parties and close cooperation between the user interface developer and the radiologist for feedback. This resulted in an iterative design method that continues even after the end of the project. All parties were actively involved in both writing and reading evaluation reports in a real-time based writing program, which has contributed to the major scientific spread that the project has resulted in.