Breast Invasive Ductal Casinoma (IDC) Detection using AlexNet and ResNet

  • Hassana Abubakar Department of Biomedical Engineering, AI and Robotics Institute, Near East University, Nicosia, Mersin 10, Turkey
  • Zubaida Sa’id Ameen Department of Artificial Intelligence Engineering, Research Center for AI and IoT, AI and Robotics Institute, Near East University, Nicosia, Mersin 10, Turkey
  • Chadi Altrjman Department of Artificial Intelligence Engineering, Research Center for AI and IoT, AI and Robotics Institute, Near East University, Nicosia, Mersin 10, Turkey
  • Sinem Alturjman Department of Artificial Intelligence Engineering, Research Center for AI and IoT, AI and Robotics Institute, Near East University, Nicosia, Mersin 10, Turkey
  • Auwalu Saleh Mubarak Department of Artificial Intelligence Engineering, Research Center for AI and IoT, AI and Robotics Institute, Near East University, Nicosia, Mersin 10, Turkey
  • Fadi Al-Turjman Department of Artificial Intelligence Engineering, Research Center for AI and IoT, AI and Robotics Institute, Near East University, Nicosia, Mersin 10, Turkey
Keywords: Cancer, Invasive Ductal Casinoma, detection, Alexnet, Resnet-101

Abstract

Cancer is one of the major health conditions worldwide having breast cancer as the most common form in women and second in rank in terms of a high death rate. IDC is the most prevalent form in which the cancer cells proliferates in ducts, invade breast fatty tissue and spread to other parts of the body with time. Therefore, regular screening for breast cancer is very important for early detection. Previous methods of cancer detection such as Nottingham and Bloom-Richardson have been used which is stressful and time-consuming. Hence, computer-aided methods are required to overcome these challenges. In this study, AlexNet and ResNet-101 models were employed for the classification of IDC. Different dataset split ratios and epochs were employed to evaluate the optimum performance of these models in this study. The optimum performances of the models were attained at 40 epochs of both 70:30 split and 80:20 split. These models’ accuracy, sensitivity and specificity were greater than 90% at 40 epochs of both 70:30 and 80:20 split which outperformed most of the previous methods of IDC detection.

Published
2022-02-02