Despite all the recent progress made in cancer diagnostics and outcome prediction, histological analysis of tumour tissue still remains to have a key role. Manual histological evaluation performed by the pathologists is, however, time-consuming and prone to subjective assessment.
In his thesis, to be examined publically 24 August, M.Sc. (Tech) Riku Turkki tackled this challenge by developing computer vision aided processes and testing their applicability for clinical purposes. The main aim of his dissertation was to investigate the use of computer vision for tissue characterisation and patient outcome prediction in cancer.
Riku has done his thesis work at FIMM in the group of Research Director Johan Lundin with Dr. Nina Linder as a co-supervisor.
He graduated from the University of Oulu in 2012, majoring in electrical engineering. Towards the end of his studies he gained deep interest in computer vision and decided to do his Master’s work in a group in Oulu focusing on texture analysis and facial recognition. The group had just started collaborating with Johan Lundin and Riku ended up working with this new and exciting medical image project.
After graduating, the existing collaboration with Lundin group brought Riku to FIMM and a year later he decided to officially start his Doctoral thesis project at the University of Helsinki.
Combining the best from the both worlds
Riku’s thesis consists of two articles and one manuscript currently being under review. All evaluate the applicability of texture analysis and/or neural networks based methods for detection of clinically relevant features from digitized cancer tissue samples with basic staining. Furthermore, in each work, the performance of the machine is compared to human expert assessment.
I have been lucky to be able to combine two very fascinating research fields. I enjoy computer vision related approaches and utilising these approaches for improving cancer care has indeed felt motivating and meaningful.
- Riku Turkki
In the first publication, Riku developed a method for quantification of tumor viability through segmentation of necrotic and viable tissue compartments. He developed a regional texture analysis method which was able to discriminate between viable and non-viable tissue regions with high accuracy when compared to human expert assessment.
The second publication focused on quantifying tumor-infiltrating lymphocytes in breast cancer specimens. Pre-trained convolutional neural networks were shown to be suitable for analysis of histological image data and to outperform texture analysis methods.
In the last publication, Riku studied computerized patient outcome prediction in breast cancer. He trained the machine with data from close to 900 patients from a nationwide patient cohort with long-term follow-up information, aiming to develop an outcome classifier with an ability to classify the patients into low or high risk groups. Testing the classifier in further 400 samples showed that prognostic discrimination was achievable without human guidance in the training of the algorithm.
We were positively surprised to see that the computerised risk prediction was also an independent prognostic factor that provided information complementary to the standard clinicopathological factors. Unfortunately identifying the features that provide this additional prognostic information is still quite challenging.
Computer vision may very well provide superhuman skills for future researchers and clinicians, once large enough clinical datasets become available.
Riku is pleased with the interdisciplinary and productive working atmosphere at FIMM and he feels that it has motivated also him to give his best. He likes to test his limits also out of working hours by hiking, ski touring and climbing in the nature, especially in the Alps.
Riku will turn a new page in his career right after his thesis defense. He will start his post-doctoral career in the group of Prof. Olli Kallioniemi at the Science for Life Laboratory in Stockholm.
M.Sc. (Tech) Riku Turkki will defend his doctoral dissertation entitled "Computer vision for tissue characterization and outcome prediction in cancer" in the Faculty of Medicine, University of Helsinki, on 24 August 2018 at 12:00. The public examination will take place at the following address: Päärakennus, Auditorium XV, Unioninkatu 34. Docent Pekka Ruusuvuori, University of Tampere, will serve as the opponent, and Professor Sampsa Hautaniemi as the custos.
The dissertation is also available in electronic form through the E-thesis service.