DeepDetect: A Novel Approach To Detecting Breast Cancer using Artificial Intelligence

In the United States, breast cancer causes about 42,000 deaths yearly. Early detection of breast cancer can boost the five-year survival rate to $93 \%$. Current breast cancer detection methods include mammograms, Computed Tomography (CT), and Magnetic Resonance Imaging (MRI) tests; breast tissue biopsy of the tumor is used to confirm the presence of cancer. Review of biopsy slides by pathologists can be fraught with human error decreasing the breast cancer diagnosis accuracy. DeepDetect is an innovative machine-learning method that can help improve the accuracy of breast cancer diagnosis from tissue biopsy samples, provide results rapidly, and help those without access to proper resources. A five-layer convolutional neural network model was constructed, optimized, and trained with 2,480 benign images and $\mathbf{5, 4 2 9}$ malignant histopathology images from the BreakHis Dataset to categorize the tissue slides into benign or malignant cell types. An attention module was added so that DeepDetect focuses on the most crucial features of the tissue images when predicting tumor state, increasing accuracy by an additional $10 \%$. With this architecture, DeepDetect achieved a final accuracy of $99.01 \%$ in diagnosing breast cancer. Compared with previous deep learning models regarding precision, recall, accuracy, loss, and F1 score, DeepDetect surpassed these models with $99 \%$, $99 \%$, $99 \%$, $0.024 \%$, and $99 \%$. Future approaches involve training the model on more diversified datasets, utilizing parallel processing, and incorporating the ability of the model to detect other cancer types.