Medical image fusion framework for radiological diagnosis

Abstract

There has been considerable advancement in the field of Digital Image Processing with wide spread applications in general and medical domain. The brain abnormalities can be captured from various discernments. There is increased dependency of the medical fraternity on use of Neuroimages obtained from different modalities. Each modality is limited in its ability to capture the complete diagnostic information. Repeated scans with multitude of data are often proposed for delineating the physiological differences, record disease progression and intent for early diagnosis. In the proposed work medical images are obtained from Med Harvard Brain Atlas http://med.harvard.edu/AALIN/home.html. The response of each modality in correlation during fusion governs the quality of the fused image. Single modal (same modality) and multimodal image (anatomical and functional modality) fusion framework are devised. The clinical brain analysis cases of degeneration and Neoplastic disease are selected for execution. Contrast Limited Adaptive Histogram Equalization (CLAHE) technique is used to augment the contrast. The multi resolution fusion framework uses Stationary Wavelet Transform with restrictive coefficient spreading by limiting the decomposition to the first level. In order to accurately transmit the salient features into the fused image hybrid fusion rules of Principal Component Analysis and absolute maximum selection are engaged. The proposed method efficiently reveals by objective assessment the imperative diagnostics details and compares the fundamental nature of single modal and multimodal image fusion. The attained value for single modal image This dwells with prudence to pass forward the salient features unaltered. Improved visibility is attained pre-processing the images to aid in better clinical interpretations by physicians. The results obtained have been compared with the state of art techniques and are found superior in quality of fused image with improved salient feature extraction for improved diagnosis.