Predicting Breast Cancer Risk from Histopathology Images using Hybrid Deep Learning Classifier

Abstract

Millions of people per year pass away from breast cancer (BC), which is a fatal illness. To deal with this issue more effectively, diagnoses can be made more scalable and less prone to mistakes by developing automated malignant BC detection systems applied on patient's imaging. Not less significant is the fact that this type of research can be expanded to include additional cancer types, having a greater influence on helping to save lives. Recent BC recognition results demonstrate that Convolution Neural Networks (CNN) can outperform manually crafted feature descriptors in terms of recognition rates. However, this higher recognition rate comes at the cost of a more complex system to develop, one that requires more time for training and specialized knowledge to fine-tune the CNN's architecture. The comprehensive understanding of the structural and cellular attributes of tissue samples that histopathological image analysis imparts is regarded as a crucial component in the rapid detection and diagnosis of breast cancer. By means of this analysis, pathologists are capable of discerning aberrant cellular proliferation, ascertaining the existence of malignant tumours, and evaluating the degree of cancer advancement. The utilization of sophisticated imaging technologies and computational algorithms in histopathological image analysis improves the precision and effectiveness of diagnosis, allowing medical practitioners to expeditiously commence suitable therapeutic strategies. In conclusion, this instrument makes a substantial contribution towards enhancing the efficacy of breast cancer management in the field of clinical practice. However, because to its ineffectiveness, the identification of breast cancer is still an unresolved problem in medical image analysis. By combining deep learning and machine learning techniques, we were able to create a categorization system built on histology images that significantly improved the accuracy of early breast cancer diagnosis while relieving doctors of some of their effort. Multilayer Perceptron and LightGBM classifiers are used in this study's analysis of histology images of breast cancer (BC). Histopathology images are the gold standard when it comes to making a diagnosis of breast cancer. Here, a dataset of 3104 photos is used to train our model, and the image is subsequently successfully identified. This strategy also yields impressive analysis and findings. We test the suggested strategy using histology images from the Breast Cancer dataset, and we achieve a high classification accuracy of 98.28%. The results of the experiments demonstrate that our strategy can perform pretty favourably and exceed cutting-edge approaches.In the near future, using deep learning to predict breast cancer may show to be quite successful.