Using parallel pre-trained types of DCNN model to predict breast cancer with color normalization

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Breast cancer is a leading cause of death in women, necessitating accurate and timely diagnosis.
• Manual pathological diagnosis is time-consuming, requires expert consensus, and is prone to errors.
• Automated expert systems can enhance diagnostic quality and efficiency in pathology.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (DCNN) model for classifying breast cancer anatomy.
• To improve the accuracy and efficiency of breast cancer diagnosis through an automated system.

Main Methods:

• Utilized the Vahadane algorithm for color staining of histopathological images.
• Developed a parallel DCNN model combining three pre-trained networks: Xception, NASNet, and Inception_Resnet_V2.
• Aggregated features from the three DCNN branches to leverage their combined strengths.

Main Results:

• The proposed DCNN model achieved 98% accuracy on the BreaKHis dataset.
• The model's performance surpassed that of other existing models in breast cancer classification.
• Performance was evaluated across various threshold ratios.

Conclusions:

• The developed DCNN model shows significant potential for accurate and efficient breast cancer diagnosis.
• This automated approach can assist pathologists, reduce diagnostic errors, and improve patient care.
• Further validation and comparison with other models confirm its efficacy.