An Efficient Cancer Classification Model for CT/MRI/PET Fused Images

The aim was to study image fusion-based cancer classification models used to diagnose cancer and assess medical problems in earlier stages that help doctors or health care professionals to make the treatment plan accordingly.

In this work, a novel image fusion method based on Curvelet transform is developed. CT and PET scan images of benign type tumors were fused together using the proposed fusion algorithm and the same way, MRI and PET scan images of malignant type tumors were fused together to achieve the combined benefits of individual imaging techniques. Then, the marker-controlled watershed algorithm was applied on fused images to segment cancer affected area. The various color features, shape features and texture-based features were extracted from the segmented image. Following this, a data set was formed with various features, given as input to different classifiers namely neural network classifier, Random forest classifier, and K-NN classifier to determine the nature of cancer. The results of the classifier showed normal, benign or malignant category of cancer.

The performance of the proposed fusion algorithm was compared with the existing fusion techniques based on the parameters PSNR, SSIM, Entropy, Mean and Standard Deviation. Curvelet transform based fusion method performs better than already existing methods in terms of five parameters. The performances of the classifiers were evaluated using three parameters: accuracy, sensitivity, and specificity. The K-NN Classifier performed better compared to the other two classifiers and it provided an overall accuracy of 94%, sensitivity of 88% and specificity of 84%.

The proposed Curvelet transform based image fusion method combined with the KNN classifier provides better results compared to other two classifiers when two input images were used individually.