DWNNet-Therm: A Deep Wavelet Neural Network Architecture Dedicated to Multiclass Classification in Breast Thermography

• Heuryk Wylk Éboli UPE
• João Freire Abramowicz UPE
• Maíra Araújo de Santana UFPE
• Wellington Pinheiro dos Santos UPE

Resumo

Early diagnosis of breast cancer is essential to increase the chances of successful treatment and improve patient survival. Thermography is a non-invasive technique that records infrared radiation emitted by the skin, enabling the detection of thermal changes associated with pathological processes, such as breast lesions. This work proposes a method based on a Deep Wavelet Neural Network (DWNN) for the automatic classification of thermographic images into four clinical categories: cyst, benign lesion, malignant lesion, and no lesion. The approach combines the multiscale analysis provided by wavelet transform with the deep learning capabilities of convolutional neural networks, allowing efficient and robust extraction of relevant thermal features for diagnosis. The model was trained and validated using a specific thermographic database. The results obtained demonstrate an overall accuracy of 74.41% in the test set, with satisfactory performance in precision, recall, and F1-score metrics for the different classes. The high precision in identifying malignant lesions (0.84) stands out, although the recall for this class was moderate (0.67), indicating the need for improvements to reduce false negatives. In conclusion, the proposed methodology presents itself as a promising tool to assist healthcare professionals in the early diagnosis of breast cancer through thermography, contributing to the development of non-invasive, accessible, and effective methods.