Development of an Optical Breast Cancer Diagnosis System Using Laser Speckle and Machine Learning-Assisted Fusion of Texture Maps

Breast cancer remains one of the most prevalent diagnosed cancers that represents a serious threat to public health. It is associated with many aggressive pathological features and lower survival rates, especially in young women. The unique absorption and scattering properties of the different constituents of breast tissue give rise to the idea of using light as a noninvasive method for identifying breast lesions. In this study, we introduce a low-cost and nondestructive optical diagnosis system based on laser speckles for the early detection of breast cancer. The proposed optical system is implemented using two independent low-power laser sources operating at 532 and 632 nm to generate sets of speckle patterns from ex vivo breast tissue samples. We then present a novel feature extraction method to capture any structure modifications caused by breast masses from such information-rich patterns. This method proposes texture map analysis based on multi-neighborhood local entropy and Gabor filter bank. To assess the discriminatory power of the extracted features, three independent supervised classification models are utilized. The experimental results indicate that features extracted from speckle patterns generated at 632 nm present higher performance than those built with 532 $\mathrm{nm}$. The merged features from both laser radiations provide a comprehensive assessment of the breast tissue characteristics. The proposed method demonstrated an enhanced performance of classification models, with accuracy values reaching up to 98.48% and weighted F1 scores up to 98.54%. This study highlights the potential of laser speckle imaging combined with AI for the early identification of breast abnormalities.