Phd in PHYSICS

Thesis title: Improvements and deep learning applications in 19F-NMR

A promising new therapeutic approach for pancreatic cancer is Proton Boron Fusion Therapy (PBFT) which produces a highly localized damaging action through nuclear reactions of the incoming proton beam and boron atoms, conveniently administered to the patient before the treatment. Starting from this recent proposal to use boron (and possibly fluorine) as chemical radiosensitizing agents in proton therapy, a new interest has arisen for the study of borate compounds. To evaluate the effectiveness of these compounds it is necessary to measure the bio-distribution of tracers accumulated in the tissues before the irradiation on a patient by patient basis. It’s safe to assume that fluorine/boron mediated sensitization will depend critically from compound concentration that can be achieved in the target nuclei which means that the clinical application of the treatment will need the development of a reliable quantification technique optimized for the tracer of interest. The first Chapter will report our study of the intracellular internalization of fluoroboron-phenylalanine (F-BPA) ,one of the most promising candidate to be adopted in PBFT as boron carrier, in a cellular model of the pancreatic cancer (PANC-1 cell line) using fluorine magnetic resonance spectroscopy (19F-MRS). The main advantage of F-BPA over the standard molecule adopted in the field, the boron-phenylalanine (BPA), is the addition of the fluorine atom that allows its quantification with magnetic resonance and it may also be used as a tracer for magnetic resonance imaging (MRI). This is the first step to validate a boron carrier as a proton therapy enhancer since PBFT damage is highly localized and its effects depend on the intracellular concentration of boron. In the second Chapter, I will discuss the possibility of measuring fluorine accumulation in tissues using 19F-MRS ex vivo in an animal model of pancreatic adenocarcinoma. This experiment will help define the sensitivity that has to be reached to perform an in vivo experiment of localized 19F-MRS and 19F MRI and will provide the data for the validation of these in vivo techniques. We also believe that this method of quantification ex vivo may be of general interest to screen for fluorine tagged compounds for the utilization in PBFT. The isotope 19F is characterized by 100% natural abundance, high relative sensitivity, it displays an intense nuclear magnetic resonance signal and it is almost nonexistent in the human body. In contrast to NMR techniques based on proton resonance, all the signal detected can be attributed to the tracer introduced and the signal that can be obtained is limited by the tracer concentration in tissues that in turn is constrained by the safety of the dosage administered and the method used for drug delivery. So, in the context of 19F-MRI, in presence of low signal and no fluorine induced background, it is extremely important to develop tools to remove noise (denoising). Thus, the third Chapter is a preliminary work on the application of a deep learning convolutional neural network (CNN) to the task of noise reduction in magnetic resonance imaging (MRI). MRI acquisition is performed in the frequency domain, I will show how a newly proposed CNN trained on raw frequency data may outperform a network of the same complexity that is trained in a more conventional way on the reconstructed magnitude images. The last Chapter consists in an application of this proposed method to a denoising task of a large dataset of parallel imaging to show how the method can be easily transferred to many other acquisition modalities.