Breast Cancer Diagnosis in Full Field PET: A Successful Method Using Two Full and Cropped Detection Paths and An Optimal Faster R -CNN

Abstract

Medically, it is crucial for discovering breast cancer in its earliest stages order to lower mortality. Two different kinds of PET tumorsmass and calcification can be classified as benign or malignant using a new computer-aided detection (CAD) and classification technique. This study examines possibilities for diagnosing and categorising breast lesions by integrating multi-modality radiomics data from positron emission tomography (PET) and magnetic resonance (MR) images to characterise breast cancer phenotype and prognosis. an OFRCNN system that locates lesions in complete and cropped PET images and then classifies themselves to determine their pathology type. The system combines the Reinforced Marine Predators Algorithm (RMPA) with an Optimal Faster Region with Convolutional Neural Networks (OFRCNN) to determine the great values for the hyperparameters of the FRCNN structure. The full-field digital PET scans from the QIN-Breast dataset are used to apply the three steps of the suggested technique. The PET images are first cleaned up in advance to get rid of any extraneous artefacts, and then they are cut into thin, overlapping slices. Second, after establishing the OFRCNN model, masses are located using two different methods: full PET detecting and cropped slice detection. The outcomes demonstrated that the OFRCNN performed better than its competitors.