Patch-based lesion detection using deep learning method on small mammography dataset

Traditional Computer Aided Detection (CAD) systems employed scanned films (low image quality) and were constructed with restricted computing resources, leading to a less reliable application procedure in breast cancer imaging. The issue of a small dataset is handled in this study using two strategies: i) using image patches as inputs rather than recognizing full-sized images; and (ii) employing the idea of transfer learning, which uses the skills learned during training for one task for another task that is closely related (also known as domain adaptation). In this regard, the CNN model that was recently trained is modified to identify masses in FFDM first, and then the CNN model that was previously taught to differentiate between mass and non-mass image patches in the Screen-Film Mammogram (SFM). Private datasets are utilized for this. The most prevalent type of cancer among women is breast cancer. According to estimates, 12% of women in the United States will receive a breast cancer diagnosis at some point in their lives [1]. Other studies have found that, except for melanoma skin cancer, breast cancer has the highest incidence and mortality rates of any type of cancer [2]. The World Health Organization's most recent statistics show that the incidence of breast cancer was the highest among oncological diseases in Uzbekistan in 2018 with a death rate of 1,449, or 0.92% of all deaths [2,3]. In 1993, there were 5.3 cases of this illness per 100,000 persons, but by 1998, that number had risen to 6.1. Every year, more people with breast cancer are receiving a diagnosis. If we take the Samarkand region alone, in 2000, 154 patients with this disease were registered in the primary register, and 2019, their number reached 300. The incidence of breast cancer has increased dramatically in recent years. Most patients are referred to doctors in stages III–IV after a long delay in the establishment of screening programs, which leads to early cancer identification, which can be credited to this. The death rate from breast cancer is the focus of intense efforts. Knowing that the stage of cancer at which it is identified affects the likelihood of survival makes it important to catch the disease early rather than later when it is more difficult to treat.

Перевод пока недоступен