In this study, we analyzed data obtained from the curative online adaptive radiation therapy of 8 patients treated between June 10, 2022, and April 14, 2023. The treatments were performed using the Ethos Therapy system (Varian Medical System, Palo Alto, USA), with cone-beam CT (CBCT)-based planning supported by artificial intelligence (AI).

The basic characteristics of the participants are summarized in Table 1.

According to our institutional treatment protocol, 3 patients received a dose of 63 Gy in 1.8-Gy fractions five times a week with conventionally fractionated radiation therapy. Another 3 patients, following tumor bed marking with lipiodol, received 63 Gy/2.1 Gy to the tumor bed and 57 Gy/1.9 Gy to the entire bladder using simultaneous integrated boost technique. Each of these patients also underwent at least three cycles of weekly concomitant 40 mg/m2 CDDP chemotherapy (3–5 cycles) during the radiation therapy course. One patient, considering his age, underwent hypofractionated radiation therapy without chemotherapy, receiving a dose of 55 Gy in 20 fractions, according to a moderately hypofractionated bladder irradiation protocol [10]. Another patient had previously undergone pelvic radiation therapy; therefore, he received 50 Gy in 2-Gy fractions to the bladder. For patients under the age of 70 (six cases), elective pelvic lymph node region irradiation was also performed.

The adaptation process begins with the initial adaptation CBCT scan. After that, quality assessment of the images is performed. In cases where factors such as gas artifacts, a filled bladder, or rectal distension are present and may complicate the planning process, appropriate interventions can be undertaken. Afterward, the contours of the influencer structures (organs that are near the target volume and whose shape and localization can influence the contours of the target volumes) are automatically generated by the Ethos AI system. During adaptive bladder radiotherapy, these influencer structures are none other than the rectum and the bladder itself. Checking the contours of the influencer structures is very important, as the planning system uses them to create a structure-guided deformable image registration (DIR) between the planning CT scans and the adaptation CBCT scans. Based on the original planning CT contours, the gross tumor volume (GTV) and CTV are propagated from the planning CT to the CBCT using DIR. These are then checked by radiotherapy technologists (RTT) and radiation oncologists together. After this, using elastic DIR, the system creates the contours of the organs at risk (OARs). Once the contours have been created and checked, the system generates a synthetic CT by deforming the planning CT based on the daily CBCT into the current geometry. Cone-beam CT images cannot be used directly for dose calculation due to their lower contrast-to-noise ratios, issues with the conversion from Hounsfield units (HUs) to tissue density, and increased motion artifacts [11]. Therefore, this synthetic CT applies the HUs of the planning CT to provide the density information for dose calculations performed in the real treatment geometry.

The planning system then generates two plans. One is based on the planning CT and the original plan but recalculated to the anatomy of the day: the scheduled plan. Another plan is made based on the new contours, considering the preset clinical goals: the adaptive plan. The clinician can decide which plan should be delivered on the day. The plans undergo a quality assurance check using the Mobius 3D System (Varian Medical Systems, Palo Alto, CA, USA). After the adaptation procedure, a second (verification) CBCT is done to check the inter-adaptation changes before the treatment [12]. Additionally, a third (post-treatment) CBCT is acquired for the first three fractions for the calculation of the patient-specific margin. Post-treatment CBCT is prepared also once a week during the whole therapy, immediately following the treatment, which allows us to detect intrafractional changes in the anatomy during the treatment and to verify that the target volume is still covered by the PTV. This online adaptive workflow is summarized in Fig. 1.

Workflow of the online adaptive re-optimization process, CBCT-cone beam CT, OAR- organs at risk

As the adaptive treatment planning is based on CBCT and gas artifacts can seriously deteriorate the visibility of organs on CBCT, online adaptive treatment can be complicated or should even be cancelled in some cases. To ensure the safety and effectiveness of adaptive radiotherapy, patients are provided with extensive pretreatment guidance. All treatments are performed with an empty bladder, and patients are required to adhere to strict fluid intake protocols. Specifically, patients should avoid consuming fluids, coffee, or tea starting 1 h prior to the treatment session. For those undergoing concurrent chemotherapy, radiotherapy is administered first to avoid rapid bladder filling and increased urine excretion.

During the treatment course, patients are advised to avoid consuming hard-to-digest or spicy foods. To minimize gas formation, medications such as Simethicone are recommended, beginning at the planning CT stage. If necessary, laxatives are prescribed to maintain optimal image quality throughout the treatment process.

A total of 239 treatment fractions with 496 CBCT images were collected during the investigation. The CBCT images were imported into the Varian Eclipse treatment planning system (TPS) and analyzed retrospectively. At least two CBCTs were performed for each fraction: an adaptation and a verification CBCT. Any anatomical differences detected during this period are referred to as adaptation changes.

During the study, bladder contouring was performed for each fraction and for each CBCT. According to our protocol for nonadaptive radiation therapy (non-ART) of the bladder, the CTV corresponds to the external contour of the bladder and the macroscopically visible tumor mass plus a 3-mm safety margin, due to the tumor cell spread outside the bladder. For CTV–PTV expansion, an isotropic 1.5-cm margin was determined. The GTV is only contoured when the tumor bed is marked with lipiodol. Following this, the GTV–CTV expansion is 3 mm, and the CTV–PTV expansion for the tumor bed is 8 mm.

In the case of online adaptive radiation therapy (oART), the initial determination CTV is the same as for non-ART treatments. After the initial experiences with oART, it became evident that a new approach was needed for CTV–PTV expansion. Each treatment is based on the contours and plans adjusted to daily anatomy; also, the variations between fractions do not need to be considered. However, considering the increased time required for the adaptation process, a patient-specific CTV–PTV margin focuses on adaptation changes.

For the CBCT images of the first three fractions, the contours of the bladder are determined offline. These bladder volumes are merged, and a surrogate PTV is created, expanding the CTV on the first CBCT so that it should cover the union of the bladder volumes. This process is performed for all three fractions, and the largest expansions in all six directions are selected to obtain the patient-specific CTV–PTV margin for the rest of the fractions. Bladder contours drawn on CBCTs 1,2, and 3 with the union bladder contour and creation of the patient-specific PTV margin are shown in Fig. 2.

Creating a patient-specific margin. a,b Bladder contours: yellow CBCT 1, orange CBCT 2, blue CBCT 3. c,d Green union of bladder contours, purple contour with patient-specific margin

The volumes of PTVs created for non-ART and oART were determined per fraction and compared with each other. Two clinically important volumes were determined. The part of the PTV that extends beyond the bladder on the verification CT scan is identified as redundant volume (rV), representing healthy tissue that is unnecessarily irradiated.If the bladder exceeded the PTV in any direction on the verification CT scan, we called it the “missed target volume” (mV). Only volumes above 1 cm3 were registered. To assess the accuracy of radiation treatment, we further investigated the relationship between rV and PTV. Differences between treatment types for PTV and rV were analyzed using mixed-effect repeated-measures analysis of variance (ANOVA).