Automated and robust organ segmentation for 3D-based internal dose calculation

Abstract Purpose In this work, we address image segmentation in the scope of dosimetry using deep learning and make three main contributions: (a) to extend optimize architecture an existing convolutional neural network (CNN) order obtain a fast, robust accurate computed tomography (CT)-based organ method for kidneys livers; (b) train CNN with inhomogeneous set CT scans validate daily dosimetry; (c) evaluate results obtained automated comparison manual done by two independent experts. Methods We adapted performant approach CT-images delineate boundaries sufficiently high accuracy adequate processing time. The segmented organs were consequently used as binary masks further convolution point spread function retrieve activity values from quantitatively reconstructed SPECT images “volumetric”/3D dosimetry. resulting activities perform calculations source organs. Results computational expense algorithm was sufficient clinical routine, required minimum pre-processing performed acceptable Dice coefficient $$93\%$$ 93% liver $$94\%$$ xmlns:mml="http://www.w3.org/1998/Math/MathML">94% kidney segmentation, respectively. addition, self-absorbed doses calculated differed $$7\%$$ xmlns:mml="http://www.w3.org/1998/Math/MathML">7% medical physicists 8 patients. Conclusion proposed may accelerate volumetric molecular radiotherapy 177Lu-labelled radiopharmaceuticals such 177Lu-DOTATOC. However, even though fully methodology based on accelerates performs accuracy, it does not remove need supervision corrections experts, mostly due misalignments co-registration between images. Trial registration EudraCT, 2016-001897-13. Registered 26.04.2016, www.clinicaltrialsregister.eu/ctr-search/search?query=2016-001897-13 .