An Improved Machine Learning and Deep Learning based Breast Cancer Detection using Thermographic Images

Abstract

New statistics show that about 950,000 individuals a year lose their lives to breast carcinoma, making it the deadliest form of cancer worldwide. Early detection and accurate diagnosis of an illness can increase the likelihood of a favorable result, decreasing the mortality rate. Premature victims can be spared if the disease is detected early. Investigators who study cancer face various challenges, including the difficulty of differentiating benign and malignant tumors and the difficulty of classifying mild and metastatic breast cancer. Here, the pattern-recognition algorithms are used to accurately identify every tumor. However, they are all based on the idea of "binary grouping" (malignant and benign). This research made use of different pre-trained networks to speed up the training process. While evaluating the efficacy of the models, many measures were used. By using the images captured during the modeling process, both breasts can be shown together, without having to split them to show the right and left sides. After the data preparation phases, the Ensemble learning model showed the highest classification performance with an accuracy value of 100% when compared to the other pre-trained network models. In this research, thermographic images of breast cancer were used to classify the disease, and the results were combined with experimental data to form a computer-aided diagnosis method. Clinical data is used to validate the models' predictions. After constructing and evaluating two models with distinct designs, the model using the same design performed best, suggesting that the clinical data decisions were essential in improving the model's performance.