Scientists have developed a ground-breaking AI tool to assist neurosurgeons in deciphering the genetic make-up of brain tumours during surgery.
This state-of-the-art technology offers vital insights into the molecular characteristics of a tumour. It enables surgeons to make real-time decisions about the precise amount of brain tissue to remove and whether it is appropriate to administer targeted anti-tumour medications to the brain.
The research report, led by experts from Harvard Medical School, was published in the Journal Med earlier this month (July 2023).
Accurate molecular diagnosis is critical in brain tumour surgery.
Removing excessive tissue in a less aggressive tumour can adversely affect neurological and cognitive functions. Conversely, insufficient tissue removal in a highly aggressive tumour may leave behind malignant cells that can rapidly grow and spread.
Now this new process can immediately instruct surgeons rather than them waiting days or weeks to learn the molecular type.
The significant milestone could eradicate the issues surrounding the profiling of tumours molecularly that have typically been a challenge in clinical practice.
The AI tool – CHARM (Cryosection Histopathology Assessment and Review Machine) – extracts biomedical signals from frozen pathology slides and provides intraoperative molecular diagnosis in real-time.
Kun-Hsing Yu, senior author and assistant professor of biomedical informatics at the Blavatnik Institute at Harvard Medical School, said: ‘Right now, even state-of-the-art clinical practice cannot profile tumours molecularly during surgery. Our tool overcomes this challenge by extracting thus far untapped biomedical signals from frozen pathology slides.’
Hailed as an advancement of real-time precision oncology, the developers suggest one of the most significant advantages is the potential for immediate treatment.
Certain tumours can benefit from on-the-spot treatment using drug-coated wafers. A Gliadel Wafer is a form of carmustine contained in a wafer. The wafer has a coating that dissolves slowly to release the carmustine directly into the area where the brain tumour was and is often placed directly into the brain at the time of the operation.
The current standard approach for intraoperative diagnosis involves a frozen section. However, freezing tissue can alter the appearance of cells and interfere with the accuracy of clinical evaluation. Additionally, even with microscopes, the human eye may not reliably detect subtle genomic variations on a slide.
CHARM’s AI approach overcomes these limitations to offer an efficient alternative.
Researchers utilised brain tumour samples from 1,524 individuals with glioma, the most aggressive and common form of brain cancer. CHARM demonstrated a remarkable accuracy of 93% when distinguishing tumours with specific molecular mutations and successfully classified three types of gliomas with distinct molecular features that have different prognoses and respond differently to treatments.
CHARM captured visual characteristics of the tissue surrounding malignant cells, allowing it to identify areas with greater cellular density and increased cell death within samples.
These indicators are associated with more aggressive glioma types. The tool also pinpointed clinically significant molecular alterations in low-grade gliomas, a less aggressive glioma subtype with different growth and treatment response propensities.
In addition to identifying molecular markers, the CHARM tool connected the visual characteristics of cells, such as the shape of their nuclei and the presence of oedema with the tumour's molecular profile.
This ability to assess the broader context around the image significantly enhances the accuracy, making it closely resemble how a human pathologist would visually assess a tumour sample.
While the CHARM tool was trained and tested on glioma samples, researchers believe it can be developed to identify other subtypes of brain cancer.
AI models have already been used to profile other types of cancer, such as colon, lung and breast cancers. However, gliomas have presented unique challenges due to their molecular complexity and the substantial variation in the shape and appearance of tumour cells.
It is worth noting that the CHARM tool would need periodic retraining to adapt to new disease classifications as knowledge and understanding of brain cancer continue to evolve.
The AI tool now requires clinical validation and clearance from regulatory bodies.
The potential of this technology to revolutionise brain tumour surgery and improve patient outcomes is highly promising, researchers say.
