Evaluation of Algebraic Reconstruction Technique algorithm for microwave imaging

Cancer is one of the leading cause of death worldwide, and an early detected cancer is likely to get better treatment. Widely used modalities for scanning the presence of cancer such as Computed Tomography and Magnetic Resonance Imaging still have problems related to the cost, size and equipment complexity. Microwave imaging is considered as an alternative modality due to its low health risk, low cost, ease of use, and portability. It exploits dielectric characteristics differences of various types of body tissue and uses tomography for image reconstruction. Tomography refers to collecting transmission or reflection data by illuminating an object from different directions to obtain its cross-sectional image. There are two major categories for image reconstruction methods, analytical reconstruction and iterative reconstruction. A widely used analytical reconstruction method, Filtered Back-Projection (FBP), has efficient computation and numerical stability. However, iterative reconstruction method such as Algebraic Reconstruction Technique (ART) efficiently improves the quality of reconstructed images, in particular when the projected data is noisy or in a limited amount. In this project, a microwave imaging system for scanning the presence of cancer in the body is designed. For obtaining the projected data, the system uses a pair of dipole antennas as a transmitter and a receiver at frequency of 3 GHz with translation of the receiver antenna, and rotation of the antenna pair (Tx-Rx) during the scanning process. Two types of numerical phantom are examined for validating the Algebraic Reconstruction Technique (ART) algorithm of the image reconstruction system. The first phantom is a 180 × 180 pixels of Shepp-Logan phantom generated from Matlab, and the second phantom is a two-layer cylindrical phantom. The cylindrical phantom has a diameter of 18 cm for the outer layer with relative permittivity is set by 4.3, representing normal tissue, and a diameter of 7.5 cm for the inner layer with the relative permittivity is set by 10.2, representing benign or malignant tissue. The simulation on quantitative parameters such as mean-squared error and structural similarity index of the reconstructed image show converging trends within 100 iterations. The value of the mean-squared error at 100th iteration is 45.16% for the Shepp-Logan phantom and 36.56% for the cylindrical phantom. The second quantitative parameter evaluated is the structural similarity index, which reaches 0.04 for the Shepp-Logan phantom and 0.008 for the cylindrical phantom.