Advanced Optical Imaging and Machine Learning for Breast Cancer Diagnosis: Wavelength Differentiation and Clinical Applications

Abstract

The rapid and accurate breast cancer (BC) diagnosis is crucial for improving patient outcomes and therapeutic efficiency. In this study, we developed an optical imaging system that integrating a spectral sensor (Neo Spectra-Micro) based on Fourier Transform Infrared (FT-IR) technology and monolithic Micro-Electro-Mechanical Systems (MEMS) to differentiate between BC from normal tissue. The system captures diffuse reflection (R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d</inf>) signals from both tissue samples, which are then analyzed to identify significant differences in R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d</inf> using a one-way Analysis of Variance (ANOVA) test, identifying 1722 nm as the most significant wavelength for differentiation (p =5.21 × 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−11</sup>). Multiple machine learning (ML) models, including Support Vector Machine (SVM), Random Forest (RF), Extra Trees (Ex. Tr), K-Neighbors (KNN), Decision Tree (DT), Gaussian Naive Bayes (GNB), Bernoulli Naive Bayes (BNB), and Logistic Regression (LR), were employed to ensure a robust and comprehensive analysis. ML-based feature selection confirmed 1722 nm as a critical predictor, aligning with findings from other Near-Infrared (NIR) studies. Performance evaluation demonstrated that DT, RF, and Ex. Tr models achieved the highest accuracy and F1 scores on independent samples. This study highlights the potential of integrating optical detection with ML for enhancing the accuracy and efficiency of BC diagnosis in clinical settings, offering a promising tool for improved cancer detection and patient care.