Zeitschrift für Stammzellforschung und -therapie

Abstrakt

Convolutional Neural Network Based Classification of Benign and Malignant Tumors from Breast Ultrasound Images

Telagarapu Prabhakar

The widely used method for diagnosing the breast cancer is a Breast ultrasound (BUS) imaging, but the interpretation
will be vary based upon the experience of radiologist. Now a days CAD systems are available to provide the information
regarding BUS image classification. However, most of the CAD systems was based upon handcrafted features. Which
are designed for classifying the tumors. Therefore, the capability of these features will decide the CAD system accuracy
which is used for classifying the tumors as benign and malignant. With the use of Convolutional Neural Network
(CNN) technology, we can improve the classification of BUS images. Because it provides a new approach for classification
and generalizable image representations thereby we can get best accuracy as a result. But, the database of BUS
image having small size so it might be restricted due to that CNNs cannot trained from scratch. To overwhelmed this
drawback, we examine the use transfer learning approach, for enabling the CNN approach to achieve best accuracy
regarding BUS image classification. The final results of VGG16_TL methodology beats the AlexNet_TL. And the final
results indicate VGG16_TL with accuracy, sensitivity, specificity, precision and F1 values of 88.23%, 88.89%, 88.89,
90% and 88.2% respectively. Therefore, we can say that possibility of pre-trained CNN models achieves good accuracy
in BUS image classification.