A Hybrid Deep Learning and Machine Learning Framework for Breast Cancer Detection in Infrared Thermography

Abstract

Breast cancer (BC) continues to be a major cause of death among women globally, highlighting the importance of early and precise detection. Infrared (IR) thermography has emerged as a promising, non-invasive, free from radiation, and economically efficient screening method by detecting temperature variations associated with malignant tissues. This study analyzes BC detection using IR thermography, utilizing a dataset of 1,520 thermal images (760 benign, 760 malignant) sourced from Kaggle. reprocessing steps included image resizing, normalization, and data augmentation was employed to enhance the model’s generalization capability. Feature extraction techniques included Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), and Gray-Level Co-Occurrence Matrix (GLCM) captured texture and structural information. Statistical features (mean, standard deviation, skewness, and kurtosis) were computed and combined, followed by Principal Component Analysis (PCA) to reduce dimensionality while preserving 95% of variance. Several machine learning (ML) classifiers Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN)—were optimized using GridSearchCV, and an ensemble model was developed through majority voting. Additionally, a deep learning (DL) pipeline was implemented using a pre-trained ResNet50 model as an element extractor with a custom classification head. The model was trained with the Adam optimizer (learning rate = 1e-4) and binary cross-entropy loss, with early stopping to prevent overfitting. The ResNet50-based model achieved 88.16% accuracy, 89.47% specificity, and 86.84% recall, outperforming traditional ML models. This study highlights the potential of integrating IR thermography with advanced ML and DL techniques for accurate, non-invasive breast cancer detection, offering a viable alternative to conventional diagnostic approaches.