|Coordinator||Karolinska Universitetssjukhuset - Innovationsplatsen|
|Funding from Vinnova||SEK 835 000|
|Project duration||October 2017 - November 2018|
Purpose and goal
In this project a total of 646 annotated pictures of pathological specimens were analyzed, stored in a data bank and analyzed with the SAABs (automated tumor detection) ATD image processing system to detect pathological changes. Mainly lymph nodes metastases of colorectal cancer were included. Data from the 646 pictures with 70941 annotated areas, gave a statistical value of the sensitivity in the region of 96.3 % - 96.8 %, in identifying pathological changes. The collaboration between Sweden, Hungary, United Kingdom, was one of the key factors to these exceptional good results.
Expected results and effects
Data from 646 pictures with 70941 annotated areas of interest, gave a statistical value of the sensitivity in the region of 96.3 % - 96.8 %, in identifying pathological changes. These results indicate there are good possibilities to develop a very useful decision support to be used within pathology. The number of detected false positive was somewhat higher than anticipated. SAAB Dynamic will prioritize their efforts in reducing this numbers, with more analyses of annotated pictures together with an effective “feed back loop” from the pathology and AI.
Planned approach and implementation
After receiving an approval from the ethical committee in Sweden and de-identifying the specimens, the department of clinical pathology and cytology, Karolinska university hospital together with the department of pathology, Semmelweis University, Budapest manually annotated a total of 646 pictures with 70941 annotated areas of interest, the ATD gave a statistical value of the sensitivity in the area of 96.3 % - 96.8 %, in identifying pathological changes. These pictures where transferred by using a computer system Cytomine from Budapest to SAAB Dynamics in Sweden.