Abstract: Breast cancer is a dangerous illness that mostly affects women. Increasing survival rates requires early detection of the disease. Artificial intelligence developments, especially in deep neural networks, have greatly improved breast cancer detection. The suggested LMHistNet, a hybrid convolutional neural network model intended to categorise microscopic images of breast tumour tissue acquired during excisional biopsy, is one such development. Levenberg-Marquardt optimization and asymmetric convolutions are used by LMHistNet to categorise breast cancer images into eight subclasses and a binary (benign or malignant) category. To achieve adaptive feature refinement, the model integrates a convolution block attention module. Additionally, batch normalisation is used to normalise input features and speed up convergence. Additionally, this method reduces internal covariance shifts during training. The model's convergence is further improved by the application of a hinge loss function, which makes it a useful instrument for precise breast cancer detection. By extracting various features from histopathological pictures, the LMHistNet model effectively performs binary and eight-class classifications that are both independent and dependent on magnification. LMHistNet's efficacy was shown in a study containing 7,909 histopathological pictures, of which 2,480 were benign and the remainder malignant. Using loss and accuracy curves, the model's performance was assessed at several magnifications (40X, 100X, 200X, and 400X). According to the findings, the model scored 88% accuracy, 89% precision, 88% recall, and 88% F1 score for multiclass categorization into eight subtypes. The model achieved remarkable accuracy, precision, recall, and F1 score of 99% for binary classification, which discerns between benign and malignant tissues. These results highlight the model's strong performance in correctly categorising photos of breast cancer at various magnifications and classes.
Keywords: Breast cancer, convolutional neural networks (CNN), deep learning, histopathological images, Levenberg–Marquardt, transfer learning.
|
DOI:
10.17148/IJARCCE.2024.13812
