Thesis

Artificial Intelligence and Optimization problems applied to the Interplay Effect in Proton Therapy

Details

  • Call:

    ProtoTera Call 2020

  • Academic Year:

    2020/2021

  • Supervisor:

    Pedro Teles

  • Co-Supervisor:

    João Santos

  • Host Institution:

    IPO - Porto

  • Granting Degree Institution:

    Universidade do Porto

  • Typology:

    National

  • Abstract:

    Artificial intelligence (AI) already plays an important role in our daily lives. From personal assistants like Apple’s Siri or Alexa from Amazon to real-time object detection on surveillance cameras. They are actually machine learning algorithms that aren’t getting smarter but improving and optimizing their skills through experience in this data-infused world we live in. In healthcare, AI is used as a tool to help doctors in early diagnostics, providing optimizations and automation in contour and volume delineation in cancer treatment planning systems or for remote patient monitoring and care. It is estimated that cancer accounted for 9.6 million deaths in 2018, being the second leading cause of death globally. (Cancer, n.d.). Radiotherapy plays an important role in cancer treatment being a reality for nearly 50% of cancer patients (Siddique & Chow, 2020). From imaging, to treatment planning, to radiotherapy delivery and verification, many things could go wrong and therefore need to be highly optimized. A patient usually has a Computed Tomography (CT) scan upon which a treatment plan is created. The treatment itself is only a few days later and the patient's internal organs will likely be in a different position. Breathing movements by themselves may introduce heterogeneities in absorbed dose distribution. These are known as interplay effects. It has also been reported that an interplay between organ motion and Multileaf Collimator (MLC) leaf motion sequences may lead to large daily variations in delivered dose (Court et al., 2008). Current planning systems are based on Knowledge-based automated planning (KBP) methods applying machine learning algorithms to assist physicians and planners to get the Organs At Risk (OAR) Dose Volume Histogram (DVH) optimal output and more recently to predict if the patient may benefit from proton therapy (Kiser et al., 2019). Protons are positively charged particles with a mass similar to neutrons. When travelling through cancer cells, protons deposit their energy damaging cellular DNA. Their penetration length is well known and energy dependent. The main advantage is that the energy deposition range is narrower and allows a more precise dose delivery framework reducing potential damage to surrounding healthy tissues and vital organs. Proton radiation therapy has been used since 1952 and for the last 10 years has been established as a major clinical tool for cancer treatment (Schreuder & Shamblin, 2020)protons’ dose absorption profiles, characterized by a Bragg peak, allow a higher radiation dose delivered to the tumour than conventional radiotherapy, while better sparing the healthy tissue. However, interplay effects are more prominent because protons’ pencil beam range is highly sensitive to organ/tissue motion (Inoue et al., 2016). Given that there is a vested interest in building a proton facility in Portugal, in this work, an AI application to correct interplay effects in the course of proton therapy treatment will be developed and tested against clinical data.

Completion status

  • Status:

  • Started At:

  • Ended At: