Cancer and cancer treatment are among today’s primary global healthcare challenges. According to the World Health Organization cancer is a second leading cause of death globally and accounted for an estimated 9.6 million deaths (one in six deaths) in 2018, with 17 million patients newly diagnosed with cancer that same year. To alleviate the alarming status and prospects, Slovenia has committed to constructing a proton therapy center by 2023, therefore it is of even more importance to advance the know-how and collaborations with innovations and improvements in cancer treatment.
More than 50% of cancer patients are treated with radiotherapy (RT), with the aim to deliver a high therapeutic dose across several daily fractions to the tumor, while minimizing exposure to the surrounding healthy tissue. While tumor localization and patient positioning can be accurately established using image-guided RT in the treatment room, accounting for day-to-day anatomical changes is difficult. Traditional approach is by inflating tumor safety margins, but at the expense of increased exposure of organs-at-risk (OAR). We focus on online adaptive RT (ART) consisting of in-room CBCT imaging for patient positioning, quantitative verification of the current plan, potential re-planning and quality assessment, all performed within 5 minutes, while the patient is immobilized in the treatment room. The immediate benefits are improved quality of RT for a wide range of disease sites, reduction of margin, OAR exposure and toxicity and/or facilitating dose escalation and improved tumor control.
However, implementation of online ART into the clinical workflow remains extremely challenging. Besides compressed timeline, the absence of integrated RT planning, imaging and delivery systems, and the limited interoperability of commercial RT planning systems and equipment seem to be the reasons why online ART is performed in only few RT sites worldwide (about 6%).
In this proposal, we aim to bridge the major research and translational gaps by developing and validating a complete suite of advanced computational tools that are required to setup the time-constrained online ART workflow in a clinical environment. The proposed project has 11 deliverables, which involve (D1) acquisition and annotation pre-treatment planning CT/MR, pre-fraction CBCT and inter-fraction CT/MR scans and (D2) collection of associated clinical outcome measures, e.g. symptoms, toxicity, recurrence and survival. For the online ART to proceed we will develop (D3) novel automated rigid CT-MR and CT-CBCT registration and (D4) deformable CT-CBCT registration, based on genetic optimizer refined by reinforcement learning. This is followed by CT to CBCT synthesis and its (D5) autosegmentation to extract tumor and OAR contours, using recent advances in end-to-end deep learning models. Next, we will develop (D6) novel and improved ART decision-support prediction models and establish relevant trigger variables and associated thresholds, sourcing from the geometric and dosimetric features, obtained by the deformable CT-CBCT registration and (D7) a fast dose calculation algorithm, respectively. If adaptation is triggered (D8) the novel re-planning pipeline will include deep learning based dose prediction and inverse treatment plan generation and optimization. Besides (D11) expected impactful research publications, the two ultimate goals are (D9) to develop and prospectively validate a treatment planning pipeline and, finally, (D10) to integrate the developed tools into a fully functional online ART workflow and perform its validation in a prospective clinical study on head&neck and prostate cancer patients at two distinct RT sites, involving photon and proton based treatment.
Project feasibility seems warranted by the strong partners that bring together four key disciplines: clinical and industrial know-how on all aspects of RT, plus expertise and experience on medical imaging and computational sciences.