Enhanced Deep Convolutional Neural Network for Breast Cancer Recurrence Prognosis

Introduction

Worldwide, breast cancer is among the leading causes of death for women. Early detection is essential for increasing survival rates. As a result, a great deal of research is being done to enhance the early identification of these tumors utilizing already accessible technologies, such as generic Machine Learning (ML) (Andrushia, A. D. et al. 2023) and other image processing approaches. Although the 5-year survival rate for breast cancer is good, metastasis is often involved and recurrence is prevalent (20% to 30% depending on stage). One problem in cancer monitoring is appropriately assigning patients to risk groups based on their initial diagnosis in order to determine the optimum course of therapy and follow-up (including risk of recurrence). To improve patient monitoring, provide high-quality treatment, and make better use of the medical resources at hand, risk categorization is essential. However, the stated accuracy for a large number of this research was often below the intended level. A promising strategy for early breast cancer detection is the Deep Learning (DL)-based approach. Mammography and ultrasonography together are a highly efficient way to find malignant breast tumors early on. MRI is also used sometimes. Qualified radiologists are required to interpret the screening results.

Cancer death rates have decreased as a result of recent advancements in earlier detection and improved treatment choices. A person's family life, profession, way of life, and health are all impacted by a breast cancer diagnosis. It entails the danger of long-lasting psychological and physical repercussions in addition to a significant mortality and morbidity risk. This sickness may also result in financial difficulties due to missed pay and medical costs. Recurrences of early-stage breast cancer in women are possible locally, regionally, or internationally. Despite the fact that 80% of these instances occur within five years of diagnosis, 30% of individuals relapse within 10 years. Differentiating between individuals who will and won't have a recurrence is presently difficult. Figure 1 shows the likelihood of a distant recurrence after years of occurrence.

Figure 1.

Chance of distant recurrence over time

In recent years, the use of DNN for medical data categorization has grown in popularity. High accuracy has been shown by these networks in applications including signal processing, picture interpretation, and clinical decision making. DNNs (Roy, A. et al. 2023) have become a popular tool for medical practitioners and academics because to its capacity to recognize intricate patterns and correlations in data. Although the creation of these networks requires a significant amount of specialized knowledge and annotated medical data, the potential advantages in terms of better patient outcomes, diagnosis, and treatment make the investment worthwhile. Drug design, testing, diagnosis, prognosis, and other medical disciplines are among the issues that DL approaches are being developed to solve. Particularly for the diagnosis and prognosis of breast cancer, DL techniques are employed. Figure 2 shows a generic DL framework for medical data categorization.

Figure 2.

Deep learning model for medical data classification

Review Of Literature

(Roy, A. et al. 2023) developed a Natural Language Processing (NLP) system to retrieve vital information regarding breast cancer from medical data. They brought these components together to produce a breast cancer medical vocabulary. They used several Machine Learning (ML) algorithms to the gathered data in order to predict the patients' risk of recurrent breast cancer. The correctness of the data was verified throughout the construction of the medical dictionary by particular users (researchers, doctors). Using this lexicon may help with personalized medication. Out of all the machine learning algorithms, the OneR algorithm had the highest sensitivity to specificity ratio.