Raman Spectroscopy-based Breast Cancer Detection Using Self-Constructing Neural Networks

Accurate and early diagnosis of cancer is an important issue in modern healthcare systems. Raman spectroscopy, as a non-invasive optical technique for evaluating intact tissues at a molecular level, has attracted the researchers’ attention. Despite recent advances, efforts are still being made to improve the sensitivity and specificity of Raman spectroscopy-based cancer detection. The present study aimed to identify three classes of breast tissues, that is, normal tissues, benign lesions, and cancer tissues, using an artificial neural network (ANN).

To improve the ANN discrimination power, a novel topologically optimized ANN, known as self-constructing neural network (SCNN), was developed in this study. The ant colony optimization algorithm was applied to optimize the topology of the network. The results of SCNN were compared with the conventional ANN, that is, multilayer perceptron (MLP).

Based on the results, the developed SCNN showed a classification accuracy of 95%.

In this study, a novel neural network (SCNN) was proposed, which was topologically optimized to improve the discrimination power of ANNs. The SCNN accuracy was determined to be 95% in Raman spectroscopy-based breast cancer diagnosis.