Hybrid Statistical and Texture Features with DenseNet 121 for Breast Cancer Classification
Unknown authors
Abstract
Cancers are aggressive which is coupled with higher mortality rates and remain a major life-threatening factor in humans.Thus, early detection of cancer among patients is important as it showed a great survival chance.Therefore, early detection provides the patients with greater survival chances.The diagnosis performed for the mammography images are quite expensive and also radiations produced during the process were harmful to the patients.Thermography is a cost-effective and invasive method compared to mammography images and thus has reached its popularity.The present research is aimed to create the Machine Learning (ML) models using Convolutional Neural Networks (CNN) approaches which were created based on the machine learning models.The present research work utilized the DMR-IR dataset for the results evaluation and performance evaluation of the model which has been verified with datasets.The feature extraction process utilizes a machine learning algorithm to overcome the problems.The developed models were engaged in sophisticated ways to extract the features to improve the classification of the model.The Hybrid statistical and texture feature extraction technique extracts the features better in turn improved the model training.The results showed that the proposed hybrid feature extraction with the Dense Net 121 model obtained better accuracy of 98.97 %, the precision of 99.45%, Recall of 98.35%, F-score of 96.85%, Sensitivity of 99.4%, and Specificity of 97.98 % when compared to the existing Multi-input CNN that obtained accuracy of 93.8%, precision of 94.1%, Recall of 97.7%, F-score of 91.4%, Sensitivity of 88.9%, and specificity of 96.7%.