BU-DLNet: Breast Ultrasonography-Based Cancer Detection Using Deep-Learning Network Selection and Feature...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early breast cancer (BC) detection and lesion characterization are vital for patient prognosis.
• Breast ultrasonography (BU) is a key radiological tool for BC diagnosis.

Purpose of the Study:

• To develop and evaluate a BU image-based framework for diagnosing breast cancer in women.
• To optimize deep feature selection and network architecture for improved diagnostic accuracy.

Main Methods:

• Utilized pre-trained deep learning networks for feature extraction from BU images.
• Employed ten wrapper-based optimization algorithms, including the Equilibrium Optimizer (EO), to select optimal deep features.
• Implemented a network selection algorithm to identify the best-performing pre-trained network.
• Classified lesions using a Support Vector Machine (SVM) classifier.

Main Results:

• The Equilibrium Optimizer (EO) algorithm demonstrated superior performance across all pre-trained models.
• Achieved a highest classification accuracy of 96.79% using ResNet-50 with a 562-feature vector.
• Inception-ResNet-v2 achieved the second-highest accuracy of 96.15% when combined with the EO algorithm.
• Results were benchmarked against existing literature findings.

Conclusions:

• The proposed BU image-based framework effectively aids in breast cancer diagnosis.
• The EO algorithm is highly effective for optimal deep feature selection in BC detection.
• This approach offers a promising tool for enhancing the accuracy and efficiency of breast cancer diagnosis.