Convolutional Neural Networks for Static and Dynamic Breast Infrared Imaging Classification

Combining Genetic Algorithms and SVM for Breast Cancer Diagnosis Using Infrared Thermography

“…In other words, our work identifies cancer cases (diagnosis task), while Tello-Mijares et al [25] only identify the presence of abnormalities in the breast (screening task). Furthermore, in general, our performance measures are very close to the performance obtained by Baffa and Lattari [23], except in values of SPEC.…”

“…[20,21], Fernández-Ovies et al [24], Sánchez-Ruiz et al [28], and Silva et al [22,27] do not use the F1-score to assess the classifiers' performance. On the other hand, only Baffa and Lattari [23] and Tello-Mijares et al [25] present an F1-score higher than ours, but both use CNN. However, it is worth highlighting that our work diagnoses cancer, while Tello-Mijares et al [25] do screening.…”

“…Baffa and Lattari [23], Fernández-Ovies et al [24], and Tello-Mijares et al [25] use a Convolutional Neural Network to classify patients into healthy and unhealthy using thermal images from the DMR-IR database [26]. Baffa and Lattari [23] used DIT (for final diagnosis) and SIT (for initial screening of cases) and obtained 98% of average accuracy with DIT images and 95% with SIT images. Fernández-Ovies et al [24] obtained best performance with Restnet34 and Resnet50, reaching a prediction of 100%.…”

“…Regarding the ACC measure, Krawczyk et al [14], Raghavendra et al [19], Santana et al [20], Baffa and Lattari [23], Fernández-Ovies et al [24], Tello-Mijares et al [25], Sánchez-Ruiz et al [28], Mishra and Rath [29], and Silva et al [27] achieve ACC values greater than 90%. Analyzing Santana et al [20], Fernández-Ovies et al [24], and Silva et al [22], we realized that they present ACC equal to 95.56%, 100.0%, and 87.96%, but they use an unbalanced dataset, and the ACC is not a good measure of performance in this case [39].…”

A Systematic Review of Breast Cancer Detection Using Thermography and Neural Networks

“…Results produced a competitive area under the curve (AUC = 0.989) of the receiver operating characteristic (ROC) curve. In [43], researchers analyzed infrared thermography of breast, considering distinct protocols, to classify patients images as healthy or nonhealthy due to anomalies such as cancer. Belongs to DMR" or belongs to the Database for Mastology Research (DMR)This dataset comprises static and dynamic protocols, with respect to their heat transfer.…”