Clinical Feasibility of Daily Online Adaptive Bladder Cancer Radiotherapy with Cone Beam CT, Using Fiducial Makers

Abstract

Radiotherapy (RT) for muscle invasive bladder cancer is challenging due to varying bladder filling. We assessed the efficacy and feasibility of online adaptive RT (oART), applying a focal boost to the tumor, in terms of dose and workflow. Bladder cancer patients (N = 15) were treated with oART on a ring-shaped Linac. This system integrates imaging (CBCT) with AI-based organ and tumor segmentation, adaptive treatment planning and delivery. Before treatment the GTV was demarcated with liquid markers. On the planning CT organs-at-risk and the GTV were contoured. The reference treatment plan was optimized with total dose for PTV (elective bladder, lymph nodes) 40Gy/20 fractions and an integrated focal boost to the GTV (15Gy). Margins were 3mm (GTV-CTV) and 5mm (CTV-PTV). Before each daily treatment, a CBCT was acquired. Bladder, rectum and GTV were determined by the AI. Planning CT and CBCT were registered to generate other organs at risk. Subsequently, the dose of the reference plan was calculated for this anatomy (scheduled plan). An adaptive plan was generated by reoptimization. Subsequently, a second pretreatment CBCT (CBCT2) was made to verify and correct the position, when necessary. Target coverage for PTV and GTV (V95%) and dose outside the target were evaluated on CBCT2. Radiation therapists (RTTs) executed the oART workflow with medical physicists (MPs) and radiation oncologists (ROs) on call. The time (median [range]) and personnel involvement were monitored. For all adaptive plans V95%>98% for CTV and GTV (boost) volumes. For scheduled plans this was 53.5% (CTV boost) and 98.5% (bladder+lymph nodes). For adaptive vs scheduled plans, the volume of dose (40Gy) to tissue outside the PTV reduced with 150cm3(p<0. 01). Median session time (patient entering-leaving) was 32 [25-45] min for the first 5 patients and reduced to 27 [20-61] min for subsequent patients. About 30% of this time was reoptimization. AI-generated GTVs were corrected in 75% for the first 5 patients and 40% for subsequent patients (taking 5min). Fiducial markers were clearly visible on CBCTs supporting GTV localization. ROs and MPs were consulted during each first fraction (5% of total) and 12% of the remaining fractions. The adaptive procedure is well feasible in clinical practice with an RTT-only workflow. The procedure takes longer than conventional RT, with reoptimization as a main factor. Dosimetric result are clearly favorable compared to delivery of non-adaptive plans.