Article Highlights
- Limited access to healthcare resources in rural or poor communities hinders early cancer detection, emphasizing the need for cost-effective and efficient techniques.
- Diffuse reflectance spectroscopy (DRS) offers a method for discriminating between normal and abnormal tissue based on spectral data, presenting an affordable option for cancer detection.
- ML holds promise in analyzing DRS data, but faces challenges due to inaccuracies in simulated datasets. A recent study proposes a wavelength-independent regressor (WIR) model to enhance ML accuracy.
- The WIR model shows promise in improving accuracy and speed compared to traditional methods like inverse Monte Carlo simulations.
Cancer diagnosis is an essential step in saving the lives of patients afflicted with this disease. Early detection of cancer dramatically decreases the mortality rate and increases the survival rate of cancer. However, the early detection of cancer is not always possible for people, especially those who live in rural or poor communities with limited access to healthcare resources (1). As a result, the development of cost-effective, accurate, and efficient techniques that are capable at detecting various types of cancer is critical to meeting the needs of underserved communities.
One spectroscopic technique that has made significant inroads in this field is diffuse reflectance spectroscopy (DRS). DRS is an analytical technique that allows for the discrimination of normal and abnormal tissue based on spectral data, which is helpful for clinicians in observing the presence of cancerous cells (2–5). The benefits of using DRS are that it is relatively inexpensive to perform this type of analysis (1).
However, DRS data still needs to be analyzed, and that is where scientists and clinicians encounter difficulties and inefficiencies. Although inverse Monte Carlo (MCl) simulations can accurately estimate the optical properties of tissue, they are computationally intensive (1). This is where machine learning (ML) could be effective. However, ML models generally require the use of simulated data sets in the data collection process, and the issue is that these data sets are inaccurate when attempting to model errors caused by instrumentation (user errors) (1).
In a recent study published in the SPIE Journal of Biomedical Optics, associate professor Bing Yu of Marquette University led a research team in exploring how to improve ML models for analyzing DRS data (6). Their study proposed a wavelength-independent regressor (WIR) model to predict the optical properties of the DRS data (6).
