Neurosurgery is a medical field where speed and accuracy make the difference between life and death, and the correct identification of the brain tumor type helps choose the appropriate treatment – usually between the invasive excision or non-invasive chemotherapy.
A new artificial intelligence tool developed by a team of Dutch researchers in molecular medicine and oncology solves both challenges.
To understand why this is extraordinarily good news, let’s review what happens to oncology patients during the examination period.
More to read:
Cancer is getting younger: study reveals 79% rise in cases among people under 50
Once the magnetic resonance imaging or tomography confirms suspicions of cancer, surgeons normally remove a flap of skull and biopsy a small portion of brain tissue for analysis. The sample is sent to the lab, where pathologists sequence and profile the brain tissue, then and attempt to identify what kind of tumor is present – a laborious process that takes a week or longer.
At the same time, they freeze a small cross section of the sample and slice it thinly with a scalpel for a quick view under a microscope. This procedure is called quick section and lasts between 15 and 20 minutes, but it’s far less reliable than the slower method.
More to read:
Study: Cancer cases and deaths among men will surge by 2050
With the model Sturgeon, an innovative AI tool developed to assist doctors, identifying brain tumors is done faster than ever and much more accurately than traditional methods allow, according to a study published in the journal Nature.
The new technology categorizes brain tumors with a 90% average accuracy in under 25 to 50 minutes, offering surgeons vital information and buying enough time to make an informed decision while patients are still on the operating table. Basically, it is a relatively affordable device ($2,000) called nanopore sequencer, which can read strands of DNA, with boosting from advanced learning algorithm that radically speed up pediatric and adult tumor identification and the subclass categorization.
More to read:
New test based on urine samples may provide early diagnosis for cancer
The Sturgeon neural network was trained using a curriculum learning method, beginning with simple simulations and progressing to more difficult ones. The researchers used nanopore sequencing data to create a model for predicting CNS (central nervous system) tumor resection.
The dataset was divided into four folds for submodel training, validation, and score calibration, including 2,801 reference labeled methylation profiles from CNS tumor and normal tissue samples. The researchers generated 500 simulated samples from the reference dataset in the 0.6% to 14% sparsity range.
Overall, Sturgeon demonstrated high accuracy, identifying 95% of definite diagnoses in 25 minutes and achieving 86% accuracy using a higher confidence threshold. After 50 minutes of simulations, 97% of models reached accurate diagnoses with a confidence level of 0.80 out of 1.0. It has also shown effectiveness in identifying tumor types in pediatric cases within 25 to 50 minutes of sequencing, correctly subclassifying 72% of cancers within 45 minutes.
More to read:
Genetically-engineered cells prevent cancer from spreading
The team – which includes scientists from the Oncode Institute, the Center for Molecular Medicine, the Princess Máxima Center for Pediatric Oncology in Utrecht, and the Departments of Neurosurgery and Pathology of Amsterdam University Medical Centers - has been experimenting with AI to identify tumors in real time since summer 2023.
The findings suggest that machine-learned diagnosis using cost-effective intraoperative sequencing could assist neurosurgeons in making critical decisions, potentially reducing neurological complications and the need for follow-up interventions.
***
NewsCafe relies in its reporting on research papers that need to be cracked down to average understanding. Some even need to be paid for. Help us pay for science reports to get more interesting stories. Use PayPal: office[at]rudeana.com or paypal.me/newscafeeu.