AI Tool Measures Brain Surgery Changes On MRI

What was studied

Researchers developed and tested a computer program called ResectVol DL that automatically outlines the cavity left after epilepsy surgery on postoperative MRI scans. They used postoperative T1-weighted MRI scans from 120 people who had epilepsy surgery, with 72 scans used to train the program and 48 used to test it. The average age at surgery was about 31.5 years, and 57 participants were women.

They compared the program's results with manual outlines, which were treated as the reference standard. They also compared ResectVol DL with three other automated methods. To check for false-positive detections and generalizability beyond epilepsy, they also tested it on 22 healthy control scans and 20 brain tumor scans.

What they found

ResectVol DL had the best overall match to the manual outlines among the tested methods on the Dice measure. Its median Dice score was 0.925, and it performed significantly better than the other three tools on that measure after Bonferroni correction.

For volume measurements, ResectVol DL and one other tool, Auto3DSeg, were both very close to the manual measurements, with very high correlations and similar relative volume differences, with no significant difference between them on those volume-based metrics. But Auto3DSeg falsely detected surgery cavities in 3 of 22 healthy control scans, while ResectVol DL and DeepResection did not show any false-positive cavities in the healthy controls tested.

The authors conclude that ResectVol DL provides accurate, fully automated segmentation of postoperative resection cavities and may be useful as a methodological tool for large-scale postoperative imaging studies in epilepsy surgery.

Limits of the evidence

This study used 120 epilepsy surgery cases, plus additional healthy control and brain tumor scans, and the abstract does not describe testing across multiple centers, scanners, or especially detailed patient subgroups. The abstract also does not say how well the tool worked in children specifically or across different epilepsy surgery locations.

The reference standard was manual outlining, which the abstract notes can be imprecise and time-consuming. Also, this study shows how well the tool matches manual segmentations, but it does not show that using the tool improves patient care or surgery outcomes.

The authors mention that the volume information might help future prediction models, but that use has not been validated yet.

For families and caregivers

For families, this study suggests that computers may be able to measure the area removed during epilepsy surgery automatically on MRI scans, which could help researchers study surgery results in larger groups of patients.

It does not mean the tool can predict how well a person will do after surgery, and it is not a treatment by itself. It is mainly a research and imaging measurement tool at this stage.

What to watch next

Useful next steps would include testing this tool in broader groups and examining whether its measurements help answer important clinical or research questions.

Terms in this summary

MRI

A scan that uses magnets to make detailed pictures of the brain.

segmentation

Using a person or computer program to outline a specific area on a scan.

resection cavity

The space left in the brain after tissue is removed during surgery.

deep learning

A type of artificial intelligence that learns patterns from many examples.

UNet-like model

A kind of computer model often used to find and outline structures in medical images.

Dice score

A number that shows how closely two outlines match; higher usually means better agreement.

false positive

A result that says something is there when it really is not.

correlation

A measure of how closely two sets of numbers change together.