Machine learning for diagnostic decision support in radiology
|Coordinator||Uppsala universitet - Institutionen för informationsteknologi|
|Funding from Vinnova||SEK 112 865|
|Project duration||September 2016 - February 2017|
|Venture||Individrörlighet för innovation och samhällsnytta|
Purpose and goal
Large quantities of information are created every day within the healthcare sector and it constitutes a potential goldmine of clinical knowledge. However, this goldmine is today neglected and underutilized. In this project, we developed a decision support system for medical doctors based on Machine learning. This system was employed to find differential diagnoses (diagnoses with similar symptoms) by analyzing thousands of unstructured historical cases. The differential diagnoses were presented as a list and accompanied by probabilities.
Expected results and effects
We developed a first prototype of the decision support system and presented it at the annual meeting of the Radiological Society of North America (RSNA) in December 2016. The prototype was much appreciated and was seen as an interesting support in the everyday work of medical doctors. The project also resulted in renewed collaborations between Sectra AB and hospitals in Skåne and in Östergötland as well as with the Machine learning research group at Uppsala University (UU). The prototype will be further developed and evaluated within these collaborations.
Planned approach and implementation
The project leader was relocated from UU to Sectra under the half-year long project. During the stay, the project leader helped out in developing the prototype of the decision support system together with employees at Sectra. The project leader was also assisting in some other machine learning projects. The project resulted in many important insights and revealed a number of problems which needs to be solved before the prototype can be employed everyday in a clinical environment. Additional funding has been obtained from Vinnova to conduct research related to these problems.