A random forest model to classify head and neck radiotherapy treatment plans
Kuan-Min Lin and Andrea Raith
Department of Engineering Science, University of Auckland
Radiotherapy treatment planning involves making conflicting treatment trade-offs between irradiating cancerous tissues and sparing surrounding critical organs. However, due to geometrical structural variations, assessing the quality of a plan is difficult. Plan tuning is conducted in a trial-and-error manner without knowing the improvement potential. This planning process leads to variable plan quality as well as inefficiency in producing treatment plans. This study proposes a planning tool based on random forest and data envelopment analysis (DEA) to support the planning process. DEA performs plan evaluation by comparing a plan to an ideal defined by the dataset and has been originally proposed for assessing the quality of prostate radiotherapy plans. For treatment sites with a more complicated structural relationship such as head and neck, DEA losses discrimination power due to the increased number of clinical criteria. We propose to use a random forest model to assess the clinical criteria and classify if a plan is satisfactory. Only plans deemed satisfactory are subsequently assessed for improvement potential by DEA.
This presentation is eligible for the ORSNZ Young Practitioners Prize.