Please click here to apply through our internal career site Find Jobs - Elekta.
We continually look for motivated and skilled individuals who are interested in supporting our customers – healthcare professionals who use our products to help patients and their communities.
We currently have the following opportunity available - please contact us for more details!
We don’t just build technology. We build hope for everyone dealing with cancer.
Master Thesis: Machine Learning Surrogates for Convolution-Based Dose Calculation in Leksell Gamma Knife Radiosurgery
What you’ll do at Elekta
Introduction
This project investigates machine learning surrogates for convolution-based dose calculation in Gamma Knife radiosurgery. The goal is to develop fast, accurate models that reproduce heterogeneity-corrected dose distributions, enabling real-time optimization without the computational cost of full convolutions.
Background
The Leksell Gamma Knife (LGK) is widely used for stereotactic radiosurgery of brain lesions. Treatment planning requires accurate calculation of the total dose delivered from 192 cobalt-60 sources. Clinically, two dose models are used: the TMR10 water-based model, which is fast but assumes water-equivalent tissue, and the convolution algorithm, which accounts for tissue heterogeneity (bone, air) using CT data and Monte Carlo-derived kernels. Convolution is more accurate, especially for targets near the skull or air cavities, but it is slower and requires CT imaging, whereas most LGK workflows are MRI-based.
The LGK Lightning inverse planner optimizes plans using the water model for speed, then optionally recalculates dose with convolution for accuracy. However, this post-hoc correction may reveal under- or over-dosage that cannot be easily fixed within the same workflow. A fast, accurate surrogate for convolution would enable real-time heterogeneity-aware optimization, improving both clinical accuracy and usability. Deep learning has shown promise in external beam radiotherapy dose prediction, and recent work in LGK radiosurgery demonstrates feasibility of multiple network designs for dose distribution prediction, as well as MRI-based heterogeneity correction without CT.
Problem Statement
The aim of this project is to investigate and prototype a machine learning model capable of predicting convolution-equivalent dose distributions for gamma knife treatments. The aims of the project are:
To develop a machine-learning based model to fast and reliably approximate convolution-based dose distributions using training data generated from the treatment planning system (TPS).
To assess the model’s potential to enable fast heterogeneity-aware dose prediction suitable for integration in optimization workflows.
Materials and Methods
Data source: Simulated LGK dose distributions generated by exporting per-sector/per-collimator dose components from the treatment planning system. MR images of patient geometries (skull outlines, targets) are available; CT images are not routinely available. However, this is not an issue, see the referenced material. The TPS convolution algorithm provides ground-truth labels.
Data volume: Large amounts of synthetic training data can be generated (multiple shots per patient, each with 24 sector/collimator combinations). Each 3D dose volume contains ~4 million voxels, requiring cropping to a bounding box around relevant dose regions (e.g., down to a dose cutoff), and possibly some level of compression.
Evaluation metrics: Gamma index passing rates, voxel-wise MAE/RMSE, and clinical indices (coverage, conformity index) compared to convolution TPS output.
Output
Prototype codebase in Python (e.g., PyTorch), including data preprocessing, baseline and extended ML models, and evaluation scripts.
Trained models and results, with benchmark metrics and visualizations (2D/3D isodose comparisons).
Master’s thesis report, including literature review, methods, experiments, and critical evaluation.
Possibly a simple demo script/notebook for dose prediction on new cases, plus presentation materials (poster or slides).
The right stuff
Curious and driven with a problem-solving mindset
Structured, independent, and proactive in your approach to research
Communicative and collaborative
Excited about working in a multidisciplinary and international team
What you bring
We are looking for students from Physics, Mathematics, Computer Science, or Engineering programs with similar orientation. Applicants should preferably have:
A good understanding of deep learning, preferably including computer vision. Basic radiation physics is an advantage.
Solid Python programming skills
Strong analytical skills
The ability to work independently
Clear communication skills, both in writing and presentations
What you’ll get
In this role, you will work for a higher purpose: hope—for everyone dealing with cancer, and for everyone regardless of where in the world, to have access to the best cancer care. You’ll join a global leader in precision radiation therapy and contribute to real-world research with direct clinical application. You'll also benefit from mentorship by experts in medical physics, machine learning, and radiosurgery.
Hiring process
We are looking forward to hearing from you! Apply by submitting your CV, academic transcript, and a short cover letter in English via the “Apply” button. Please note that we do not accept applications by e-mail.
Your Elekta contact
For questions, please contact the Global Talent Acquisition Partner responsible, Gustaf Ericson, at gustaf.ericson@elekta.com. Again, we do not accept applications through e-mail.
We are an equal opportunity employer
We are an equal opportunity employer. We evaluate qualified applicants without regard to age, race, colour, religion, sex, sexual orientation, gender identity, genetic information, national origin, disability, veteran status, or any other protected characteristic.
About Elekta
As a leader in precision radiation therapy, Elekta is committed to ensuring every patient has access to the best cancer care possible. Elekta is a proud innovator and supplier of equipment and software used to improve, prolong, and save the lives of people with cancer and brain disorders.
More than 6,000 hospitals worldwide rely on Elekta technology. We openly collaborate with customers to advance sustainable, outcome-driven, and cost-efficient solutions to meet evolving patient needs, improve lives and bring hope to everyone dealing with cancer. To us, it's personal, and our global team of 4,700 employees combine passion, science, and imagination to profoundly change cancer care. We don’t just build technology, we build hope.
Elekta is headquartered in Stockholm, Sweden, with presence in more than 120 countries and listed on Nasdaq Stockholm. For more information, visit elekta.com or follow @Elekta on LinkedIn and Twitter.