Skip to main content
placeholder image

Breast Mass Tumor Classification using Deep Learning

Conference Paper


Abstract


  • © 2020 IEEE. This study presents pre-trained Convolutional Neural Network (CNN) models to classify pre-segmented mammogram mass tumors as benign or malignant. Transfer learning, particular pre-processing and data augmentation were applied to overcome the limitation of the scarcity of available training dataset. The proposed models are based on modified versions of Inception V3 and ResNet50 to tackle the classification problem mentioned above. The proposed architectures have been tested on the Digital Database for Screening Mammography (DDSM) dataset, and it achieved an accuracy of 0.796, precision of 0.754, and a recall of 0.891 on InceptionV3-like CNN model. On the other hand, an accuracy of 0.857, precision of 0.857, and a recall rate of 0.873 have been achieved with the ResNet50-like CNN network. Overall, the proposed ResNet50-like model achieved a 5% improvement in accuracy compared to the existing state-of-the-art method for this dataset.

Authors


  •   Abdel Rahman, Anas (external author)
  •   Belhaouari, Samir (external author)
  •   Bouzerdoum, Salim
  •   Baali, Hamza (external author)
  •   Alam, Tanvir (external author)
  •   Eldaraa, Ahmed (external author)

Publication Date


  • 2020

Citation


  • A. Abdel Rahman, S. Belhaouari, A. Bouzerdoum, H. Baali, T. Alam & A. Eldaraa, "Breast Mass Tumor Classification using Deep Learning," in 2020 IEEE International Conference on Informatics, IoT, and Enabling Technologies, ICIoT 2020, 2020, pp. 271-276.

Scopus Eid


  • 2-s2.0-85085497566

Ro Metadata Url


  • http://ro.uow.edu.au/eispapers1/4049

Start Page


  • 271

End Page


  • 276

Abstract


  • © 2020 IEEE. This study presents pre-trained Convolutional Neural Network (CNN) models to classify pre-segmented mammogram mass tumors as benign or malignant. Transfer learning, particular pre-processing and data augmentation were applied to overcome the limitation of the scarcity of available training dataset. The proposed models are based on modified versions of Inception V3 and ResNet50 to tackle the classification problem mentioned above. The proposed architectures have been tested on the Digital Database for Screening Mammography (DDSM) dataset, and it achieved an accuracy of 0.796, precision of 0.754, and a recall of 0.891 on InceptionV3-like CNN model. On the other hand, an accuracy of 0.857, precision of 0.857, and a recall rate of 0.873 have been achieved with the ResNet50-like CNN network. Overall, the proposed ResNet50-like model achieved a 5% improvement in accuracy compared to the existing state-of-the-art method for this dataset.

Authors


  •   Abdel Rahman, Anas (external author)
  •   Belhaouari, Samir (external author)
  •   Bouzerdoum, Salim
  •   Baali, Hamza (external author)
  •   Alam, Tanvir (external author)
  •   Eldaraa, Ahmed (external author)

Publication Date


  • 2020

Citation


  • A. Abdel Rahman, S. Belhaouari, A. Bouzerdoum, H. Baali, T. Alam & A. Eldaraa, "Breast Mass Tumor Classification using Deep Learning," in 2020 IEEE International Conference on Informatics, IoT, and Enabling Technologies, ICIoT 2020, 2020, pp. 271-276.

Scopus Eid


  • 2-s2.0-85085497566

Ro Metadata Url


  • http://ro.uow.edu.au/eispapers1/4049

Start Page


  • 271

End Page


  • 276