Doctoral dissertation

Kari Antila: Computer-assisted medical image analysis methods can speed up the planning and outcome assessment of therapies

Kari Antila.
The analysis and interpretation of the contents of volumetric (three dimensional) medical images usually requires that the objects of interest in the image, such as tumors, are detected and outlined. A single volumetric image may, however consist of several hundred of image slices. The manual browsing and drawing of the objects of interest on this amount of image slices would require considerable amount of time and effort from a human expert. Thus the planning and assessment of the results of image-guided therapies will require computer-assisted, automatic, robust and accurate analysis methods.

Kari Antila will present in his dissertation work three methods that were developed for planning dental implants and treating uterine muscle tumors (fibroids). The objects of interest in the study, facial bones and tumors attached to the uterus, may take individual shapes. Also the used imaging modalities often suffer from distortions and other types of image quality degradation that may result from restrictions set by scanning time and X-ray exposure.

“An analysis method intended for clinical use should be robust and accurate despite of the challenges. The method needs also to be fast in order to minimize wait times of the implant planning workflow and to enable the outcome assessment even during the therapy to treat the muscle tumors,” says Kari Antila

An analysis problem that has the purpose of dividing an image into consistent and meaningful objects of interest is usually referred to as image segmentation. Antila in his dissertation work he first developed a segmentation method for the dental implant planning that uses a surface model pre-built from several past patients to be fitted around the mandible (jaw bone) of a new patient.

The use of case for the method was later extended to also cover the segmentation of other facial bones. This new use case required a new, data-driven method to be developed for segmenting and integrating the surfaces patches on the facial bones at the same time tolerating the image distortions and discontinuities of these surfaces. This same method was later applied to the segmentation problem of muscle tumors that were treated with magnetic resonance imaging guided high-intensity focused ultrasound.

“We were able to reach the average accuracy of 0,5 mm from the surface acquired with the developed automatic method to a surface drawn by a human expert. At the same time we were able to reach the execution time of less than a minute per image volume when comparable methods would take from tens of minutes to hours,” says Antila.

The research work behind the dissertation was funded by Business Finland, the VTT Technical Research Centre of Finland Ltd and Solita Ltd in projects Tilakuva, ATLAS, EDIFI, SalweImage and IVVES. The work was also enabled by grants from the Artturi and Aina Helenius fund of the Finnish Cultural Foundation, the Instrumentarium Science Foundation and the Student’s Union of the Southern Ostrobothnia at the University of Helsinki.

The doctoral dissertation of MSc (Tech) Kari Antila in the field of signal processing titled Volumetric Image Segmentation for Planning of Therapies will be publicly examined at the Tampere University, Faculty of Engineering and Natural Sciences at 12 o’clock on Friday the 29th of January 2021. Professor Miika Nieminen from University of Oulu will be the opponent while Professor Emerita Ulla Ruotsalainen will act as the custos.

The examination will be in Finnish and it can be only followed live online via remote connection.

The dissertation is available online at