Machine Learning Optimisation for Realistic 2D and 3D PET-CT Phantom Study
Mhd Saeed Sharif *
Department of Electronic and Computer Engineering, School of Engineering and Design, Brunel University, London, United Kingdom
Maysam Abbod
Department of Electronic and Computer Engineering, School of Engineering and Design, Brunel University, London, United Kingdom
Luke I Sonoda
Paul Strickland Scanner Centre, Mount Vernon Hospital, United Kingdom
Bal Sanghera
Paul Strickland Scanner Centre, Mount Vernon Hospital, United Kingdom
*Author to whom correspondence should be addressed.
Abstract
An experimental study using artificial neural network (ANN) is carried out to achieve the optimal network architecture for proposed positron emission tomography (PET) application. 55 experimental phantom datasets acquired under clinically realistic conditions with different 2-D and 3-D acquisitions and image reconstruction parameters along with 2min, 3min and 4min scan times
per bed are used in this study. The best scanner parameters are determined based on the ANN experimental evaluation of the proposed datasets. The analysis methodology of phantom PET data has shown promising results and can successfully classify and quantify malignant lesions in clinically realistic datasets.
Keywords: Image Analysis, Positron Emission Tomography (PET), Tumour, Segmentation, Artificial Neural Network