COMPARATIVE ASSESSMENT OF VARIANTS OF SIMULTANEOUS ALGEBRAIC RECONSTRUCTION TECHNIQUE WITH PROPOSED HYBRID FILTERED BACK PROJECTION ALGORITHM

Abstract

Introduction: This paper introduces a novel hybrid approach to computed tomography (CT) image reconstruction, designed to enhance medical imaging techniques. The study meticulously compares the performance of this innovative method with established algorithms, including back projection, simultaneous algebraic reconstruction (SAR), and simultaneous algebraic reconstruction iteration (SART) coupled with a total variation minimization algorithm. The evaluation utilizes the NIH-AAPM-Mayo Clinic CT Grand Challenge dataset, ensuring robust and relevant results. Two key performance metrics are taken for comparison: the Structural Similarity Index Measure (SSIM) and the Peak Signal-to-Noise Ratio (PSNR).

Objectives: To enhance the quality of filtered back projection (FBP) images in low-dose imaging,

Methods: A hybrid model is proposed and compared with SART variants. It combines FBP with a modified CNN featuring three 2D convolution layers (32, 64, and 128 filters of size 3x3) and a ReLU activation function, with an input shape of 150x150 in a batch size of 8. Three 2D max-pooling layers with 2x2 kernel sizes are included. The output is flattened and passed through a dense layer (128 units, ReLU activation), followed by a dropout layer (0.5) to reduce overfitting. The final dense layer uses Softmax for activation. The model is compiled with categorical cross-entropy the loss function and the ADAM as optimizer. Training occurs on a machine with an NVIDIA GEFORCE RTX GPU (6 GB memory).

Results: The hybrid algorithm achieved an SSIM value of 0.7916, indicating superior structural fidelity in reconstructed images. Additionally, it demonstrated a PSNR of 19.0424 dB, confirming its effectiveness in producing higher-quality images.

Conclusions: The findings have crucial implications for medical imaging, promoting safer practices to reduce radiation exposure by enhancing image quality. This balance between quality and safety highlights the significance of the research, which could revolutionize diagnostic methods in healthcare. Overall, it represents a pivotal advancement in hybrid CT reconstruction, paving the way for further innovations in medical imaging technologies.