Hybrid deep learning and genetic algorithms approach (HMB-DLGAHA) for the early ultrasound diagnoses of breast cancer

• Published in: Neural computing & applications Vol. 34; no. 11; pp. 8671 - 8695
• Main Authors: Balaha, Hossam Magdy, Saif, Mohamed, Tamer, Ahmed, Abdelhay, Ehab H.
• Format: Journal Article
• Language: English
• Published: London Springer London 01.06.2022; Springer Nature B.V
• Subjects: Artificial Intelligence; Breast cancer; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Datasets; Deep learning; Experiments; Genetic algorithms; Image Processing and Computer Vision; Machine learning; Optimization; Original Article; Performance measurement; Probability and Statistics in Computer Science; Recall; Regularization; Ultrasonic imaging; Genetic algorithms (GA); Convolutional neural network (CNN); Breast ultrasound images (BUSI); Deep learning (DL); Breast cancer (BC)
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-021-06851-5

Breast cancer is one of the most commonly diagnosed cancers in the world that has overtaken lung cancer and is considered a leading cause of molarity. The current study objectives are to (1) design an abstract CNN architecture named “HMB1-BUSI,” (2) suggest a hybrid deep learning and genetic algorithm approach for the learning and optimization named HMB-DLGAHA, (3) apply the transfer learning approach using pre-trained models, (4) study the effects of regularization, optimizers, dropout, and data augmentation through fourteen experiments, and (5) report the state-of-the-art performance metrics compared with other related studies and approaches. The dataset is collected and unified from two different sources (1) “Breast Ultrasound Images Dataset (Dataset BUSI)” and (2) “Breast Ultrasound Image.” The experiments implement the weighted sum (WS) method to judge the overall performance and generalization using loss, accuracy, F1-score, precision, recall, specificity, and area under curve (AUC) metrics with different ratios. MobileNet, MobileNetV2, InceptionResNetV2, DenseNet121, DenseNet169, DenseNet201, RestNet50, ResNet101, ResNet152, RestNet50V2, ResNet101V2, ResNet152V2, Xception, and VGG19 pre-trained CNN models are used in the experiments. Xception reported 85.17 % as the highest WS metric. Xception, ResNet152V2, and ResNet101V2 reported accuracy and F1-score values above 90 % . Xception, ResNet152V2, ResNet101V2, and DenseNet169 reported precision values above 90 % . Xception and ResNet152V2 reported recall values above 90 % . All models unless ResNet152, ResNet50, and ResNet101 reported specificity values above 90 % and unless ResNet152, ResNet50, ResNet101, and VGG19 reported AUC values above 90 % .