Machine Vision based Intelligent Breast Cancer Detection

Artificial intelligence, especially deep learning, has sparked a great deal of interest in bioinformatics, particularly complications in clinical imaging. It has achieved great success by helping the CAD system achieve high-precision results. Despite this, detecting breast cancer on mammography images is still considered a critical challenge. The work aims to decrease FPR and FNR and increase the value of MCC. To achieve this goal, two state-of-the-art object detection models are used, YOLOv5 and Mask RCNN.YOLOv5 detects and classifies the mass as benign or malignant. Due to the spatial limitations of YOLOV5, the original model is modified to achieve the desired results. Mask RCNN detects the edges of tumours invading the breast parenchyma and also detects the size of the tumours. The size of the tumours defines the stage of cancer.  The model was trained on the INbreast dataset with YOLOv5+Mask RCNN. The performance of the proposed model was evaluated compared to the original version of YOLOv5. The proposed technique achieves higher performance with a lower False-positive rate of 0.05 and False-negative rate of 0.03 and a high MCC value of 92.02%. The experiments performed show that the accuracy of YOLOv5 in combination with Mask RCNN is 0.06 higher than that of YOLOv5 alone. Additionally, this work could help determine the patient's prognosis and allow physicians to be more accurate and predictable at early-stage breast cancer detection.