Dual-energy CT based mass density and relative stopping power estimation for proton therapy using physics-informed deep learning

Abstract

Proton therapy requires accurate dose calculation for treatment planning to ensure the conformal doses are precisely delivered to the targets. The conversion of CT numbers to material properties is a significant source of uncertainty for dose calculation. The aim is to develop a physics-informed deep learning (PIDL) framework to derive accurate mass density and relative stopping power (RSP) maps from dual-energy computed tomography (DECT) images. The PIDL framework allows deep learning (DL) models to be trained with a physics loss function, which includes a physics model to constrain DL models. Five DL models were implemented including fully connected neural networks (FCNN), dual-FCNN (DFCNN), and three variants of ResNet: ResNet-v1 (RN-v1), ResNet-v2 (RN-v2), and dual-ResNet-v2 (DRN-v2). Two empirical DECT models were implemented to compare with the PIDL method. An artificial neural network (ANN) and the five DL models trained with and without physics loss were explored to evaluate the PIDL framework. DL training data were from CIRS electron density phantom 062M (Computerized Imaging Reference Systems, Inc., Norfolk, VA). The performance of DL models was tested by CIRS adult male, adult female, and 5-year-old child anthropomorphic phantoms. For density map inference, the physics-informed RN-v2 was 3.3%, 2.9% and 1.9% more accurate than ANN for the adult male, adult female, and child phantoms. The physics-informed DRN-v2 was 0.7%, 0.6%, and 0.8% more accurate than DRN-v2 without physics training for the three phantoms, respectfully. The results indicated that physics-informed training could reduce uncertainty when ANN/DL models without physics training were insufficient to capture data structures or derived significant errors. DL models could also achieve better image noise control compared to the empirical DECT parametric mapping methods. The proposed PIDL framework can potentially improve proton range uncertainty by offering accurate material properties conversion from DECT.

A component method to delineate surgical spine implants for proton Monte Carlo dose calculation

“…However, the super‐resolution images are compromised in patients due to limited kV/mAs and patient scatter. 23 , 24 , 31 , 34 , 38 Furthermore, the 0.24‐mm super‐resolution CT image sets are 16 times the size of a conventional image set. Adaptive CT image resolution would be encouraged for TPS vendors to allow more accurate modeling of implant materials and volume.…”

“…Screw length can also be verified with fine‐resolution CT images as the tapered threads of screws can be seen. However, the super‐resolution images are compromised in patients due to limited kV/mAs and patient scatter 23,24,31,34,38 . Furthermore, the 0.24‐mm super‐resolution CT image sets are 16 times the size of a conventional image set.…”

Prostate gleason score prediction via MRI using capsule network

“…Radiomics, the extraction of image features, has been used to build models aiding in lesion detection, lesion synthesis and cancer prognosis [9,10]. In recent years, many deep learning models have achieved state-of-the-art results in image diagnosis [11][12][13], organ segmentation [14,15] and treatment planning [16,17] in medicine. Recent advances in deep learning have also demonstrated success of convolutional neural networks (CNN) to detect cancer from MRI.…”

Ultrasound-based dominant intraprostatic lesion classification with Swin Transformer

“…Due to its feasibility in detecting DILs, B-mode ultrasound could be potentially integrated into the intraoperative treatment planning for a focal boost [2]. With the development of computer-aided diagnosis (CAD), powerful artificial intelligence (AI) tools such as deep learning (DL) and reinforcement learning algorithms have been applied in medical image analysis [3][4][5][6][7][8][9][10], such as image segmentation [11], image registration [12] and image synthesis [13]. With deeper layers, DL methods can adaptively extracts feature maps at multiple resolution levels, and demonstrates strength in computer vision [14].…”