Dosimetric comparison of the BNCT treatment planning performances when using a nnU-NET to automatically segment Glioblastoma Multiforme

Abstract

This work presents a preliminary evaluation of the use of the convolutional neural network nnU-NET to automatically contour the volume of Glioblastoma Multiforme in medical images of patients. The goal is to assist the preparation of the Treatment Planning of patients who undergo Boron Neutron Capture Therapy (BNCT). BNCT is a binary form of radiotherapy based on the selective loading of a suitable 10-boron concentration into the tumour and on subsequent low-energy neutron irradiation. The selectivity of the therapeutic effect is based on the capacity of the boron drug to target preferentially cancer cells, thus triggering the neutron capture only in the tumour and depositing there a lethal dose. Even if the tailoring of the beam to the tumour volume is less crucial for BNCT than for other radiation therapies, a proper delimitation of the tumour volume is needed to assess a safe and effective dosimetry. In clinical application the contour must be manually decided by the physician, however, a tool to automatically define important structures such as the Gross Tumour Volume (GTV) and the Organs At Risk (OAR) would be beneficial to enable medical physicists assessing preliminary positioning and simulated dosimetry before the approval or possible changes introduced by the radiotherapist. Moreover, an initial contouring may speed up the work of the physician. The nnU-NET was trained and tested and its performance was evaluated through different parameters such as the Dice Coefficient. To assess a more meaningful evaluation for BNCT, for the first time, this work analyzed the difference of the clinical dosimetry in 16 patients using the manual and the automatic contoured images.