Multileaf collimators (MLC) are an essential component in modern radiotherapy that shape the X-ray treatment beam. Currently, MLC leaf position accuracy is verified to ±1 mm every month. However, leaf position accuracy can drift between verification dates and treatment verification only occurs pre-treatment. To prevent serious errors, it would be highly beneficial to use a real time verification system. We are developing a system based on Monolithic Active Pixel Sensors (MAPS). MAPS are radiation hard under photon and electron irradiation, have high readout rates, low attenuation and are suitable for high resolution applications making them an ideal upstream radiation detector. Here, we report results using the Lassena MAPS, which measures 12×14 cm 2 and is three side buttable, allowing full treatment fields to be monitored. The Lassena detector was placed in the treatment field of an Elekta Synergy LINAC which has an MLC leaf width of 0.5 cm. An MLC leaf was extended to 10 different positions within the field. Sobel based methods were used to reconstruct the leaf edge position. Correspondence between reconstructed and set leaf position was excellent and resolutions ranged between 60.6±8 µm and 109±12 µm for a central leaf with leaf extensions ranging from 1 to 35 mm using ∼0.3 s of treatment beam time while the sensor was placed at an SSD of 85 cm.

Monolithic Active Pixel Sensor (MAPS) devices are an effective tool for upstream verification of Intensity Modulated Radiotherapy (IMRT) treatments. It is crucial to measure with high precision the positions of the multi leaf collimators (MLC) used to shape the beam in real time, in order to enhance the quality and safety of treatments. This work describes r-UNet, a deep learning based solution for leaf position reconstruction. The model is used to analyse the high-resolution images produced by a Lassena MAPS device in order to automatically determine the leaf positions. Image segmentation and leaf position estimation are performed simultaneously in a multi-task setting. r-UNet obtained an average Dice coefficient of 0.96 ± 0.03 for the reconstructed image masks in the held-out test set; whilst the mean squared error (MSE) resulting from the estimation of the MLC positions is 0.003 mm, with a resolution ranging between 45 and 53 µm for leaf extensions between 1 and 35 mm. On unseen leaf positions, r-UNet yielded a single-leaf resolution between 54 and 88 µm depending on the leaf extension, and an average MSE of 0.07 mm. These results were obtained using single frames of data collected at 34 frames per second.

The current trend in X-ray radiotherapy is to treat cancers that are in difficult locations in the body using beams with a complex intensity profile. Intensity-modulated radiotherapy (IMRT) is a treatment which improves the dose distribution to the tumor whilst reducing the dose to healthy tissue. Such treatments administer a larger dose per treatment fraction and hence require more complex methods to verify the accuracy of the treatment delivery. Measuring beam intensity fluctuations is difficult as the beam is heavily distorted after leaving the patient and transmission detectors will attenuate the beam and change the energy spectrum of the beam. Monolithic active pixel sensors (MAPSs) are ideal solid-state detectors to measure the 2-D beam profile of a radiotherapy beam upstream of the patient. MAPS sensors can be made very thin (∼30 µm) with still very good signal-to-noise performance. This means that the beam would pass through the sensor virtually undisturbed (<1% attenuation). Pixel pitches of between 2 µm to 100 µm are commercially available. Large area devices (∼15×15 cm 2) have been produced. MAPS can be made radiation hard enough to be fully functional after a large number of fractions. All this makes MAPS a very realistic transmission detector candidate for beam monitoring upstream of the patient. A remaining challenge for thin, upstream sensors is that the detectors are sensitive to the signal of both therapeutic photons and electron contamination. Here, a method is presented to distinguish between the signal due to electrons and photons and thus provide real-time dosimetric information in very thin sensors that does not require Monte Carlo simulation of each linear accelerator treatment head. Index Terms-Clinical/preclinical evaluation/application studies, dosimetry for radiation-based medical applications, monolithic active pixel sensors (MAPSs), Monte Carlo simulations for imaging and therapy, radiation detectors for medical applications, radiotherapy verification.

Higher energy and intensity X-ray radiotherapy treatments are coming into wider use, having the benefit of requiring fewer treatment fractions and fewer hospital visits per patient. However, small percentage errors in multileaf collimator positioning and dose become bigger problems with higher doses per fraction. Hence, real-time treatment verification becomes essential. Where devices downstream from the patient suffer from scattering in the patient, upstream devices can disturb the therapeutic beam. Here, a method is proposed for performing dosimetry upstream using monolithic active pixel sensors, which can be made thin enough to disturb the beam by <1%. In order to calculate the dose to the tumour, a verification device needs to make a measurement of the photon field. Some photons will Compton scatter an electron in the silicon and generate a signal. However, this signal is obscured by energy deposits from contamination electrons, originating from Compton scattering in the accelerator head and air. Often extensive build-up material is added to verification devices to reduce the electron contamination and enhance the photon signal. However, this leads to degradation of the beam intensity to the patient. Instead we propose using thin strips of 50 µm thick copper in a grating pattern and measuring the difference in the signal with and without it. The contamination electrons are relatively undisturbed by the presence of the thin copper strips and the photon signal generated via Compton scattering is enhanced under the strips. Hence the difference in the two signals mostly consists of energy deposits originating from the therapeutic photons. From this the dose to the patient can be derived. The 50 μm of copper is thin enough to keep the beam attenuation below 1%, but in itself would give a total signal response which consists of 38% contamination electrons. Using the grating technique, we show that the electron contamination can be reduced to 2.6% of the total signal. This allows the photon signal only to be extracted from the data and thus the dose to patient with a very thin upstream detector can be calculated using a Monte Carlo model to extrapolate the photon flux into the tumour.

A multileaf collimator (MLC) is an integral component in modern radiotherapy machines as it dynamically shapes the photon field used for patient treatment. Currently, the MLC leaves which collimate the treatment field are mechanically calibrated to ±1 mm every 3 months and during pre-treatment calibration are calibrated to the mechanically set leaf positions. Leaf drift can occur between calibration dates and hence exceed the ±1 mm tolerance. Pre-treatment verification, increases LINAC usage time so is seldom performed for each individual patient treatment, but instead for an acceptable sample of patients and/or treatment fractions. Independent real-time treatment verification is therefore desirable. We are developing a large area CMOS MAPS upstream of the patient to monitor MLC leaf positions for real-time treatment verification. CMOS MAPS are radiation hard for photon and electron irradiation, have high readout speeds and low attenuation which makes them an ideal upstream radiation detector for radiotherapy. Previously, we reported on leaf position reconstruction for single leaves using the Lassena, a 12 × 14 cm2, three side buttable MAPS suitable for clinical deployment. Sobel operator based methods were used for edge reconstruction. It was shown that the correspondence between reconstructed and set leaf position was excellent and resolutions ranged between 60.6 ± 8 and 109 ± 12 μm for a single central leaf with leaf extensions ranging from 1 to 35 mm using 0.3 sec of treatment beam time at 400 MU/min. Here, we report on leaf edge reconstruction using updated methods for complex leaf configurations, as occur in clinical use. Results show that leaf positions can be reconstructed with resolutions of 62 ± 6 μm for single leaves and 86 ± 16 μm for adjacent leaves at the isocenter using 0.15 sec at 400 MU/min of treatment beam. These resolutions are significantly better than current calibration standards.