Researchers have developed an innovative technology combining optical fluorescence microscopy and artificial intelligence, capable of predicting genetic mutations associated with lung cancer. This could help doctors choose the most appropriate treatment faster while preserving limited biopsy samples for other tests.

Genetic mutations in lung cancer are critical factors in determining the effectiveness of targeted therapies, making rapid detection a medical priority.

Researchers at the University of Edinburgh and the UK National Health Service said the technology was able to predict the presence of EGFR gene mutations with high accuracy, and also distinguished between two of the most common and clinically important types of these mutations for treatment decisions.

The study, published in the journal Cancer Research, a journal of the American Association for Cancer Research, noted that the technology uses so-called "fluorescence lifetime imaging microscopy," which measures the brief time it takes for natural molecules in tissues to emit light after being stimulated by a light source.

Cells and tissues produce different patterns of natural fluorescence depending on their chemical composition, metabolic activity, and disease state. A deep learning model analyzes these signals to find signatures associated with specific mutations, without the need to stain the sample or extract its DNA.

Mutation Determines Treatment Path

The epidermal growth factor receptor (EGFR) gene is one of the important genes in some cases of non-small cell lung cancer, particularly adenocarcinoma.

Some mutations in this gene can make cancer cells dependent on abnormal growth signals, allowing the use of targeted drugs that inhibit the activity of the protein produced by the gene.

For this reason, doctors need to know whether the tumor harbors a targetable mutation before choosing treatment. Failure to detect a mutation may deprive a patient of a potentially more suitable drug, while prescribing targeted therapy to a patient without the appropriate mutation may delay more effective treatment.

Currently, the process of identifying mutations relies primarily on molecular tests such as DNA sequencing. Although accurate, these tests may require specialized laboratories and time ranging from days to weeks, and they consume part of the tumor sample, which may already be small.

Preserving tissue becomes even more important when the biopsy is taken with a fine needle or from a hard-to-reach area, as doctors may need to use the same sample to confirm the tumor type, search for several other mutations, and evaluate markers that help select immunotherapy or targeted therapy.

Reading Light Instead of DNA

The new technology does not directly read the genetic sequence of the tumor, but attempts to infer its genetic status from the changes caused by the mutation inside the cells.

Cancer mutations affect processes such as energy production, metabolism, and the balance of molecules within the cell. These changes can alter how tissue components absorb and emit light, leaving a pattern that AI algorithms can learn.

The researchers placed unstained samples of lung cancer tissue under the new microscope, then trained a computer model to distinguish between tumors carrying EGFR mutations and those without.

They tested the system's ability to differentiate between exon 19 deletion and the L858R mutation in exon 21, two of the most common alterations among EGFR-related lung tumors. Distinguishing between them may be important when evaluating treatment options and expected outcomes. The data released by the researchers shows separate groups for mutated and non-mutated tissues, as well as these two subtypes.

The researchers said the model achieved high performance in identifying the mutation status and type, but they have not yet presented the technology as a ready replacement for the genetic tests used in hospitals.

Minutes Instead of Weeks

The researchers said scanning the sample could take minutes, compared to molecular processes that may take weeks, and the potential cost could drop from thousands to hundreds of British pounds if the technology proves effective in widespread clinical use.

Study co-author Qiang Wang, a researcher at the Institute for Regeneration and Repair at the University of Edinburgh, said the approach paves the way for transforming an expensive and slow process into a faster and cheaper test, especially in hospitals and health systems without easy access to complex molecular testing.

The method does not require chemical processing or permanent staining of the tissue, meaning the sample remains relatively intact and can be used later for pathological examination, gene sequencing, or other tests.

This advantage could be important as lung cancer screening programs expand, leading to the detection of more nodules and early tumors, increasing the number of biopsies that need analysis in pathology labs.

But despite the promising results, optical prediction of a mutation does not mean the system has actually read the tumor's DNA.

Metabolic or optical changes caused by different mutations may be similar, and the signals can be affected by how the biopsy was collected and stored, the type of microscope, imaging settings, and the amount of cancer cells in the sample.

An AI model may perform well within the hospital or dataset on which it was trained, but its accuracy may drop when tested on patients from other regions or on samples prepared using different devices and protocols.

Some tumors may contain more than one group of cells, each with different genetic characteristics, known as intratumor heterogeneity. This could result in a mixed optical signal that is difficult for the system to interpret.

Therefore, the researchers are currently working on clinical validation of the approach, including testing it on larger and more diverse patient groups and comparing its results directly with standard genetic tests.

If subsequent studies succeed, the technology is likely to be used first alongside molecular tests, not as a replacement, either to identify samples with a high probability of mutation, to speed up initial decisions, or to preserve tissue.

The researchers plan to test the technology on other types of cancer, search for additional mutations that can be targeted by drugs, and integrate the examination into routine laboratory workflows.

The importance of this technology lies in its ability to provide accurate genetic information without the need for complex, time-consuming analyses. If proven clinically effective, it could help accelerate treatment and improve patient outcomes. However, its widespread success requires validation through further studies involving larger samples and a broader range of mutations.