Data Collection from Thermal Imaging device in IOT

Abstract

Breast cancer, being one of the most prevalent malignancies, poses a formidable threat to human health due to its aggressive nature and elevated mortality rates. The pivotal role of early detection in augmenting patient survival rates is well-established. Presently, mammography serves as the conventional diagnostic approach; however, its cost and exposure to ionizing radiation underscore the need for alternative, cost-effective, and less invasive diagnostic modalities, such as thermography. In light of this, the goal of the current research project is to construct and create a thermal imaging-based breast cancer detection model. The first phase entails developing a customized machine learning model built on convolutional neural networks (CNNs), which is specifically intended to use thermal pictures to detect breast cancer. This model is subsequently fine-tuned through extensive training using a diverse dataset comprising thermal images of breast abnormalities, aiming to achieve a robust detection mechanism. The overarching objective of this study is to facilitate binary classification, distinguishing malignant from benign breast cancer cases, with a particular emphasis on the potential to enhance diagnostic accuracy, especially when confronted with multifarious image attributes. This work employed a wide range of image classification methods to identify breast cancer utilizing thermal image processing techniques. The comprehensive workflow encompassed cancerous image enhancement, precise segmentation, texture-based feature extraction, and the subsequent classification of breast cancer within thermal images, culminating in a successful endeavor. Concomitant with these intricate challenges, a bespoke classifier was devised, capitalizing on machine learning paradigms such as the 2D Convolutional Layer (2D CNN) and Support Vector Machine (SVM). The proposed model was meticulously trained on a representative dataset, meticulously selected from the DMR-IR (Database of Mastology Research). Notably, the empirical results yielded a classification rate of 95% for the proposed 2D CNN classifier, surpassing the SVM and pre-existing CNN counterparts, which registered classification rates of 91% and 71%, respectively. It is crucial to emphasize that there are now just a handful of publicly available datasets for thermography in the field of cancer diagnosis.