Cancer is typically diagnosed at an advanced stage when survival rates are low. Most early-stage cancers are asymptomatic, and traditional methods such as imaging or histopathological testing are not feasible as routine screening tests for the general population due to high cost and other clinical constraints. While several surface-enhanced Raman scattering (SERS)-based cancer detection methods have been developed to boast high sensitivity and selectivity, they tend to focus on a single or just a few biomarkers, and often only for a narrow range of cancer types, hampered by an insufficient sample size. Moreover, many researchers remain at the preliminary stages, lacking data that use easy to interpret and failing to leverage more efficient high-throughput analysis methods.

In a new paper published in eLight, a team of scientists, led by Professor Xiangheng Xiao from the College of Physical Sciences, Wuhan University, have taken a significant leap forward by developing a label-free SERS-Artificial intelligence method for cancer screening (SERS-AICS). 

SERS-Based Cancer Screening with AI

This technology merges the detection strengths of traditional SERS systems with the analytical power of advanced big data tools. The team tested as little as 15ul of patient serum samples with Ag nanowires each for lung, colorectal, hepatic, gastric, and esophageal cancers, capturing the subtle changes in vibrational signals of molecules in cancer samples due to their altered physiology and pathology. The researchers then trained and validated their predictive workflow to recognize cancer by analyzing the molecular vibrational spectrum from two independent cohorts involving 382 healthy individuals and 1,582 cancer patients. The system demonstrated impressive efficacy with an accuracy of 95.81%, a sensitivity of 95.40%, and a specificity of 95.87% overall for five cancer types. Additionally, it was successful in distinguishing samples at an early stage of cancer from those with common diseases, while facilitating the creation of a data platform for more in-depth analysis.

“This was very promising, as early-stage screening should detect changes in molecular fingerprinting information that are intermediate between healthy and disease states,” says Xiao. “And what’s truly exciting is that it isn’t restricted to one or a just handful biomarkers, but expands to encompass an all-inclusive ‘panoramic’ view for every single alternative signals in cancers.”

Further reading: Lung Cancer Mortality Risk Predicted by Blood Test

Xiao added: “Our study demonstrates the potential for developing a sensitive tool for the early detection of various cancers. The predictive technique can identify individuals potentially harboring cancer from their blood samples obtained in routine health exams. Anyone with suspicious findings would then be referred for further evaluation by definitive diagnosis.”

In future work, the researchers plan to analyze the spectrum of molecular vibration associated with various clinical characteristics of cancer to gain a comprehensive understanding of the disease, potentially aiding in selecting targeted therapies. They also aim to broaden the application of the SERS-AICS method to detect a wider range of cancers and other diseases, promising a transformative step forward in early-state cancer detection and patient monitoring.

Featured image: Fig 1. SERS-AICS characterization of five cancers with high mortality. ROC curves with covariance matrices-assisted SVM model for distinguishing (A) 244 lung cancer patients, (B) 216 colorectal carcinoma patients, (C) 195 gastric cancer patients, (D) 203 hepatocellular carcinoma patients, (E) 193 esophageal carcinoma patients, (F) 400 mixture cancer patients from 324 healthy controls in the internal cohort. The (G) accuracy, (H) sensitivity and (I) specificity of single or multiple cancers/healthy control, the overall accuracy, sensitivity and specificity of all cancers could reach at 95.81%, 95.87%, 95.40%. The 400 mixed cancer patients were obtained by randomly selecting 80 samples from the five types of cancer each. Photo: eLight