New ML-based tool 98 per cent effective in detecting breast cancer early: Study

New machine learning-based screening method is 98 per cent effective in detecting the earliest sign of breast cancer, according to a study.

Developed by researchers at the University of Edinburgh, the fast, non-invasive technique combines laser analysis with machine learning. It is the first of its kind to identify patients in the earliest stage of breast cancer, they said, noting that it may pave the way for a screening test for multiple forms of cancer.

The technique can pick up subtle changes that occur in the bloodstream during the initial phases of the disease—known as stage 1a—which goes undetected by existing tests.

Physical examination, X-ray or ultrasound scans, or analysis of a sample of breast tissue, known as a biopsy are standard tests currently available for breast cancer. These rely upon screening people based on their age or if they are in at-risk groups.

The pilot study, published in the Journal of Biophotonics, involved 12 samples from breast cancer patients and 12 healthy controls. In the study, the team optimised a laser analysis technique—known as Raman spectroscopy—and combined it with machine learning.

The team could spot breast cancer at stage 1a with 98 per cent effectiveness.

It first shines a laser beam into blood plasma taken from patients. Using a spectrometer device, the team analysed the properties of the light after it interacted with the blood. The spectrometer then revealed tiny changes in the chemical makeup of cells and tissues—early indicators of disease.

Using the machine learning algorithm, physicians can interpret the results. Using the novel approach, the team could also distinguish between each of the four main breast cancer subtypes with an accuracy of more than 90 per cent. The team said this enabled patients to receive more effective, personalised treatment.