Developing An Accurate Model for Comprehensive Analysis Of Mammographic Masses In Clinical Settings
Abstract
Mammography, a key tool in early threat detection, drives chest screening programs aimed at spotting infections early. These programs, managed by BI-RADS, yield vast data analyzed by radiologists. This study focuses on AI models predicting mammography outcomes, aiming to reduce unnecessary biopsies. Naïve Bayes and K-Nearest Neighbor algorithms were tested, with Naïve Bayes showing the highest accuracy at 85.43%.
Keywords
Download Options
Introduction
Breast cancer stands out as one of the most prevalent diseases affecting women. In 2016 alone, approximately 246 thousand new cases of invasive breast cancer were reported, alongside 61 thousand cases of non-invasive forms. It's an arduous journey for any cancer patient, serving as both a challenge and a constant vigil. Early detection becomes imperative due to the high mortality rates associated with advanced stages of the disease. Mammography emerges as the cornerstone in diagnosing breast cancer, recognized for its reliability and widespread use. The Breast Imaging Reporting and Data System (BI-RADS), established by the American College of Radiology, initially categorized mammogram results into four classifications, later expanded to six. Mammography is lauded for its cost-effectiveness and efficiency in identifying risks during the preclinical stage, although chest screening programs have not been fully optimized for early disease detection.
Clinical evaluation utilizing the BI-RADS scale may necessitate further biopsy before a conclusive diagnosis can be made by the expert. Biopsy results can vary, ranging from benign to malignant growths. While some biopsies may confirm benign conditions, the necessity for biopsy arises when the expert's confidence in the patient's BI-RADS assessment from the mammogram is uncertain. Alarmingly, nearly 70% of biopsies yield benign outcomes, a considerable number that could potentially have been avoided. Radiologists exhibit considerable variability in interpreting mammograms, prompting the utilization of Fine Needle Aspiration Cytology (FNAC) in such cases. However, FNAC's typical accuracy rate stands at 90%, leaving room for potential misdiagnosis.
The primary objective of the BI-RADS system is to categorize patients into those with no evidence of breast cancer (benign) and those with strong indications of malignancy, aiding in treatment decisions based on mammographic findings and patient demographics. This study aims to assess the proficiency of experts in distinguishing the severity of mammographic abnormalities based on BI-RADS classifications, biopsy outcomes, and patient age. [1]. [3]. [5]. [7].
Conclusion
This paper explores two distinct classification models, artificial neural network and support vector machine, for predicting the severity of breast masses. The proposed method addresses missing values, trains, and optimizes both models. The primary focus is on developing an accurate classification model for clinical analysis of mammographic masses. Experimental results show that the naïve Bayes model outperforms the KNN technique in terms of learning accuracy and complexity.