AI can diagnose CNS tumours during surgery

An AI algorithm called "Sturgeon" can classify CNS tumours fast and accurately during surgery. This enables resection strategies adapted to tumour types.

Uncertainty factor: exact diagnosis of CNS tumour not known before surgery

Tumours of the central nervous system (CNS) are among the deadliest types of cancer, especially in children. The primary treatment is often neurosurgical resection. The extent of the resection must be weighed against the extent of the neurological damage induced by the resection. In addition, for some tumours, such as ependymomas, complete resection improves the prognosis, whereas this is not the case for other tumours, such as medulloblastoma. Another complicating factor is that the exact tumour type is often not known before the operation. The current standard procedures - preoperative imaging and intraoperative histological analyses - are not always conclusive and are occasionally incorrect.

Tumour diagnosis thanks to rapid sequencing and AI analysis

With the help of rapid nanopore sequencing (see below), a "small" methylation profile can be created from a tumour tissue sample during the operation. The small nanopore sequencing device can be connected directly to a laptop via USB cable, which can make an accurate tumour diagnosis using the specially trained AI "Sturgeon" (Ultra-fast deep-learned CNS tumour classification during surgery; DOI: https://doi.org/10.1038/s41586-023-06615-2).

"Sturgeon" is a deep neural network that was trained using simulated nanopore sequencing data. This was generated from extensive, publicly available methylation profiles of CNS tumours and control tissues. The performance of "Sturgeon" was evaluated on a test dataset, achieving high accuracy in tumour classification after approximately 40 minutes of simulated sequencing.

"Sturgeon" AI passes test and use in the operating theatre

"Sturgeon" was very successful in the retrospective classification of 50 CNS tumour samples, providing an accurate diagnosis in 45 cases (90%). When used in real time during 25 neurosurgical operations, "Sturgeon" achieved correct classification in 18 out of 25 cases (72%), with diagnoses made in less than 90 minutes. In the remaining 7 cases, the pre-defined confidence threshold (of 95%) was not reached. It should also be noted that the small sample size of 25 intraoperative cases is not sufficient to calculate sensitivity and specificity. Further clinical validation is therefore required.

Conclusion: AI can support neurosurgical decision-making

The diagnosis made using "Sturgeon" AI is based on a cost-effective intraoperative sequencing technique. It can support neurosurgical decision-making and potentially prevent neurological comorbidities and/or avoid additional surgery.

The principle of nanopore sequencing

After DNA extraction from the tissue, the DNA liquid is pipetted directly into the nanopore sequencer. The DNA sample is transported through a nanometre-sized pore (hence the name nanopore). As it passes through, the pressure at this pore changes. This change in pressure is specific to each of the four nucleic bases, allowing the base sequence to be read. In addition to the base sequence, nanopore sequencing can also recognise methylation patterns. This is done using specialised software.