Can machine learning methods contribute as a decision support system in sequential oligometastatic radioablation therapy?

Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced AnalyticsCancer treatment is among the major medical challenges of this century. Sequential oligometastatic radio-ablation (SOMA) is a novel treatment method that aims at ablating reoccurring metastasis in a single session with a targeted high dose of radiation. To know if SOMA is the best possible treatment method for a patient, the benefits of each available therapy need to be understood and evaluated. The ability to model complex systems, such as cancer treatment, is the strength of machine learning techniques. These techniques have improved the understanding of numerous medical therapies already. In some cases, they can serve as medical support systems if they deliver reliable results that doctors can trust and understand. The results obtained from applying numerous machine learning techniques to the data of SOMA-treated patients show that there are favorable techniques in some cases. It was observed that the Random Forest algorithm proved superior at different classification tasks. Additionally, regression problems opposed a great challenge, as the amount of data is very limited. Finally, SHAP values - a novel machine learning interpretation technique – provided valuable insights into understanding the rationale of each algorithm. They proved that the machine learning algorithms could learn patterns aligned with the human intuition in the problems presented. SHAP values show great potential in bridging the gap between complex machine learning algorithms and their interpretability. They display how an algorithm learns from the data and derives results. This opens up exciting possibilities for applying machine learning algorithms in the real world