ABNORMALITY DETECTION AND CLASSIFICATION ON MAMMOGRAPHY IMAGES VIA CONVOLUTIONAL NEURAL NETWORKS

Avcı H., Durhan G., Demirkazık F., Gülsün Akpınar M., Karakaya Karabulut J.

12th International Conference of the International Biometric Society's Eastern Mediterranean Region, İzmir, Turkey, 8 - 11 May 2023, pp.69

  • Publication Type: Conference Paper / Summary Text
  • City: İzmir
  • Country: Turkey
  • Page Numbers: pp.69
  • Hacettepe University Affiliated: Yes

Abstract

ABNORMALITY DETECTION AND CLASSIFICATION ON MAMMOGRAPHY IMAGES VIA CONVOLUTIONAL NEURAL NETWORKS

Breast cancer is the most common type of cancer among women. Digital mammography screening is an imaging method that helps determine the rate of early detection of breast cancer. In the last two decades, Computer Aided Detection (CAD) systems were developed to help radiologists analyze screening mammograms. Since 2012, deep convolutional neural networks (CNN) have been shown to achieve high performances in image recognition. In this study, the effectiveness of deep learning using Convolutional Neural Networks (CNN) was tested on digital mammography images in detecting the presence of lesions (abnormal/normal). In addition, classification performances of different CNN architectures will be compared. Digital mammography images were used for this study, which received ethics committee permission from Hacettepe University. Craniocaudal (CC) view were carried out for each breast. Spyder environment (with the Keras and TensorFlow) was used for image processing steps (pre-processing, segmentation, feature selection, classification) and evaluating the performance on the classification. The data set was divided into two as training and test sets with 80% and 20%. Noise was removed from the images with image pre-processing techniques. Then, the regions of interest (ROI) obtained after segmentation were determined by CNN with linear filters and activation functions. The features were obtained from the ROIs with the help of the GLCM matrix. So far, the performance of 87 lesioned image (malign and benign images) 59 lesion less images (normal image) in the CNN model was examined. The applied CNN model reached 79.64%, 80.85% and 75.44% performance rates for accuracy, sensitivity and specificity, respectively. By increasing the number of observations, the analyzes will be renewed and also the classification performances of different CNN architectures will be compared. Keywords: Mammography classification, Deep learning, Medical imaging processing, Computeraided detectio