Home About Login Current Archives Announcements Editorial Board
Submit Now For Authors Call for Submissions Statistics Contact
Home > Archives > Volume 18, No 7 (2020) > Article

DOI: 10.14704/nq.2020.18.7.NQ20185

A Convolutional Neural Network based Feature Extractor with Discriminant Feature Score for Effective Medical Image Classification

R. Banupriya, Dr.A. Rajiv Kannan


In Computer-Aided Diagnosis (CAD) systems, major role is played by classification of medical images. Conventional methods uses texture features, color and shape information in a combined manner for classification. These methods are problem specific and in medical images, they have shown their complementary, which makes the systems inability to make high-level problem domain concepts representation and they are having worst model generalization ability. In recent days, because of its admirable performance in different fields, great attention is gained by convolutional neural networks (CNN). However, complete training of a novel deep CNN model is concentrated in recent works to target issues with restricted data and time consuming issues. Various investigations are done for rectifying those drawbacks of existing techniques. They utilized a CNN models a feature extractor for feature representation construction which is helpful for classification and they are successfully applied in remote sensing scene classification. For region classification, in order to in cooperate pre-trained CNN model’s multilayer features, fusion strategies are used in this investigations. Fully connected and diverse convolutional layers deep features are haul out by using pre trained CNN model as a feature extractor. Then, convolutional features are constructed by computing multidimensional enhanced discriminative feature score. At last, for classification, regression based kernel discriminant method is integrated with fully connected layers features and convolutional layers intermediate level features. With high-resolution remote sensing data sets, via MATLAB, proposed technique is evaluated for validation and better performance is exhibited in contrast with fine-tuning CNN models, fully trained CNN models and other related works.


Feature Fusion, Multi-dimensional Features, Convolutional Neural Networks (CNN), Discriminative Feature Score, Medical Image Classification

Full Text