Artificial intelligence can speed up the planning of radiation therapy for breast cancer

Almost all cancer patients receive radiation therapy at some point in their treatment. Treatment requires a dose plan tailored individually for each patient, specifying the precise targeting and amount of radiation. Developing this plan is a complicated and time-consuming process, which increases the need for new, more efficient solutions.

AI may help adjust radiation therapy 

The aim of Sanni Sinisalo’s master’s thesis was to develop a deep learning model. The model developed in her thesis was based on a previously developed model from Kuopio University Hospital (KYS) that predicts the dose distribution for radiation therapy in breast cancer patients. The model’s predictions were compared to clinical dose distributions used in actual clinical practice. 

- The study also evaluated the potential of an AI-based dose prediction model to accelerate or improve the dose planning process in the future, explains Sanni Sinisalo from the University of Jyväskylä.  

The model uses patients' CT scans 

The dose prediction model uses patients' computed tomography (CT) scans and the structure masks, which delineate key organs and tissues. CT scans are taken of all patients at the beginning of radiation therapy. Images serve as the basis for both clinical planning and the AI model’s predictions. 

- The model was able to predict dose distributions specifically for cancer patients with left-sided breast cancer, says Sinisalo.  

AI predictions are almost on the same level as clinical practice

The study results show that the dose distributions predicted by the AI model were very close to clinically used distributions. The differences between the predictions and the clinical results were small. Although for some individual organs, the prediction might estimate either higher or lower doses compared to the clinical results. 

Based on the results, the dose prediction model is effective in producing high-quality and realistic estimates of radiation therapy dose distributions.

Artificial intelligence is a tool for experts 

The study suggests that solutions based on deep learning could significantly support radiation therapy dose planning in the future. 

- The model can assist and potentially speed up the time-consuming planning process. It is also possible that artificial intelligence could even find more optimized solutions than humans in some cases, notes Sinisalo. 

However, the study emphasizes that the dose prediction model is theoretical and does not take into consideration all the practical restrictions with treatment planning.  

- AI cannot replace people but is a tool that experts can be used as part of the planning process, explains Sinisalo.  

The collaboration provided a unique learning environment  

The thesis was carried out in collaboration with Central Finland Hospital Nova and Kuopio University Hospital. The thesis was supervised by Arto Javanainen from the University of Jyväskylä, Katariina Näkki and Ville Raatikainen from Central Finland Hospital Nova and Henri Korkalainen from Kuopio University Hospital (KYS).  

- The collaboration between the different organizations was truly valuable. I gained both a research-based and clinical perspective. It helped me understand how AI solutions can be developed to serve practical healthcare needs, says Sinisalo.  

The master’s thesis is available online at: https://jyx.jyu.fi/jyx/Record/jyx_123456789_111524?sid=469788548

Further infotmation:

• Sanni Sinisalo, sanni.e.sinisalo@student.jyu.fi (available until 31.07.2026)
• Senior Researcher Arto Javanainen, arto.javanainen@jyu.fi, +358406146881