A Hybrid and Novel Optimization Framework for Denoising and Classification of Medical Images Using Dtcwp and Neuro-fuzzy Classifiers

Computed tomography (CT) images are usually corrupted by several noises from the measurement process complicating the automatic feature extraction and analysis of clinical data. To attain the best possible diagnosis it is very vital that medical images be clear, sharp, and free of noise and artifacts. In this research paper, we propose a robust technique to denoise, detect and classify the tumour part from CT medical images. Our proposed approach consists of four phases, such as denoising, region segmentation, feature extraction and classification. In the denoising phase Dual Tree Complex Wavelet Packets and Empirical Mode Decomposition are used for removing noise. Here, histon process is used in order to surmount the smoothing filter type and it will not affect the lower dimensions. We have taken into consideration two noises, Gaussian and salt & pepper for proposed technique. The performance of the proposed technique is assessed on the five CT images for the parameters, PSNR and SDME. In the segmentation process K-means clustering technique is employed. For the feature extraction, the parameters contrast, energy and gain are extracted. In classification, a modified technique called Cuckoo-Neuro Fuzzy (CNF) algorithm is developed and applied for detection of the tumour region. The cuckoo search algorithm is employed for training the neural network and the fuzzy rules are generated according to the weights of the training sets. Then, classification is done based on the fuzzy rules generated. From the obtained outcomes, we can conclude that the proposed denoising technique have shown better values for the SDME of 69.9798 and PSNR of 25.4193 for salt & pepper noise which is very superior compared to existing methods. Moreover our proposed technique has shown an accuracy of 96.3% which is very better than the existing methods.