Evaluation of Preprocessing and Feature Extraction Techniques for Enhanced Medical Image Fusion

Image fusion (IF) integrates complementary information from multiple modalities into a single enhanced image for improved analysis. Pre-processing and feature extraction techniques act as core components in IF framework. This research work presents an analysis of pre-processing and feature extraction techniques that can aid in an improved fusion process. The pre-processing techniques employed in this work are normalization, noise reduction, and histogram equalization, which ultimately enhance image quality for feature engineering. However, feature extraction is performed using (PCA) and Canny edge detection (CED) independently, that captures structural and edge details, respectively, for better fusion. The experiments are performed on medical imaging dataset consisting of 2283 images captured by the computed tomography modality. The proposed framework includes experiments performed in combination of each pre-processing technique with every other feature extraction technique. The performance is measured using Mean Squared Error (MSE), Mean Absoluate Error (MAE), Peak-Signal to Noise-Ratio (P SNR), Structured Similarity Indexing Method (SSIM), Jaccard Index (JI), Dice Coefficient (DC) and Nor-malized Cross-Correlation (NCC) metrics. The combination of normalization and PCA, achieves superior results with lowest MSE (0.000159), highest PSNR (48.1), SSIM (0.9952), and NCC (0.9998), demonstrating its effectiveness among all the combinations. Finally, a morphological image fusion is performed, whereby a priori combination of image normalization with PCA is performed to obtain fusion results.