DEVELOPMENT OF AUTOMATED RADIOTHERAPY TREATMENT PLANNING FOR CERVICAL AND BREAST CANCER FOR RESOURCE-CONSTRAINED CLINICS

Globally, cancer rates are on the rise, especially in low- and middle-income countries (LMICs). However, many of these countries lack access to radiotherapy, which is due in part to a substantial shortage of the staff necessary to deliver safe and effective radiotherapy. This staff shortage could be mitigated by the automation of the radiation treatment planning process. To this end, we developed automated planning for cervical and postmastectomy breast cancer radiotherapy, the two most common types of cancer in women in many LMICs. For radiotherapy of cervical cancer in resource-constrained clinics, the recommended treatment technique is a four-field box. We created algorithms to plan four-field box treatments with homogenous dose distributions by automatically determining the beam apertures and relative beam weights. Using our techniques we automatically planned 150 four-field-box treatments and 89% were scored acceptable by radiation oncologists. The dose distributions were more homogenous (p For radiotherapy of node-positive, postmastectomy breast cancer, it is recommended to treat the chest wall and ipsilateral nodes, while reducing the dose to normal tissues, such as the heart and lungs. We created algorithms to plan three-field treatments (mono-isocentric tangential and supraclavicular fields) on free-breathing patient CTs. The dose distribution was automatically optimized by using mixed energy photon beams and field-in-field dose modulation. Using these algorithms, we automatically planned radiotherapy treatments for 10 left-sided, postmastectomy patients. The plans were evaluated quantitatively based on their dose distributions, and 90% of the plans met constraints for lung dose, heart dose and target coverage. Physicians accepted all plans either as-is (50%) or with only minor changes (50%). Automatic QA of the plans flagged 92% of the changes requested by physicians. To assess the risk of failure in our automated treatment planning workflow, we performed Failure Modes and Effects Analysis (FMEA). FMEA showed that a specially-designed automated QA program reduced the risk of automated treatment planning. Additionally, we found that human error is still a prominent cause of potential failures and that manual plan reviews of automatically generated plans are still vital for safe delivery of radiotherapy. In conclusion, automated treatment planning and QA for radiotherapy of cervical and breast cancers were clinically viable for a majority of patients tested. Our algorithms will be implemented clinically at our partner hospitals in South Africa in the next year