"If we could make scalp EEG stop acting like a blurry security camera and start acting like a real map, that would be a pretty good day," is the kind of reaction you can imagine lead author Xiyuan Jiang having here. For people with drug-resistant focal epilepsy, finding the exact patch of brain tissue that keeps lighting the fuse can decide whether surgery helps and whether a patient finally gets a break from seizures.

That is the setup for a new PNAS paper from Jiang, Cai, Gonsisko, Worrell, and He, who built what they call a spatial-temporal-spectral imaging framework, or STSI for short (Jiang et al., 2025). The name sounds like it was assembled by a committee that loves hyphens, but the idea is simple: use one framework to compare different EEG seizure markers head-to-head.

The Brain’s Bad Group Chat

EEG records electrical activity from the scalp. That is useful, but the skull is a rude intermediary. Signals blur on the way out, which means doctors often know something is wrong without getting a clean answer to where the trouble started. In epilepsy surgery, that location matters a lot.

Researchers have a few EEG clues to work with. There are classic interictal spikes - brief bursts seen between seizures. There are seizures themselves, which are informative but inconvenient because you have to wait for one to happen. And there are high-frequency oscillations, or HFOs, very fast ripples in the signal that have become one of the more interesting suspects in the hunt for epileptogenic tissue. Reviews over the last few years argue that HFOs may be more specific than ordinary spikes, but they are also messy, hard to detect, and capable of being either pathological or perfectly normal depending on context (Frauscher et al., 2021; Noorlag et al., 2022).

That "depending on context" bit is the whole movie. A fast ripple in the brain is not automatically villain-coded.

One Framework to Rule the Noise

The STSI system tries to sort that mess by looking at brain activity across space, time, and frequency together. The team analyzed 2,081 events from 42 patients with focal drug-resistant epilepsy and checked their source estimates against clinical ground truth, including intracranial EEG seizure-onset zones and surgical outcomes.

The ranking was the interesting part. Seizures localized best overall, with an average localization error of 6.67 mm in seizure-free patients. But among the signals you can catch between seizures, the winner was not the ordinary spike and not the generic HFO. It was the HFO that overlaps with a spike - what the paper calls pHFO. Those landed at 8.73 mm on average, better than HFO-riding spikes at 10.28 mm, much better than general spikes at 19.59 mm, and miles ahead of generic HFOs at 36.53 mm (Jiang et al., 2025).

That result fits with where the field has been drifting. A 2024 Epilepsia paper from some of the same group already showed that spikes carrying concurrent HFOs line up with clinical ground truth better than conventional spike imaging (Gonsisko et al., 2024, PMCID: PMC11647446). The broader message is almost insultingly human: context matters.

Why Anyone Outside a Lab Should Care

If these results keep holding up, the practical payoff is obvious. Better noninvasive mapping could help teams narrow where to place intracranial electrodes and improve confidence about what tissue is actually generating seizures. That would matter especially in hard cases where MRI is unrevealing or several candidate regions are all acting suspicious, like a room full of people suddenly interested in the floor.

It also addresses a very current problem in epilepsy care: surgery can be life-changing, but the path to it is often long, invasive, and uncertain. Reviews keep making the same point - EEG source imaging is promising, yet real-world adoption depends on accuracy, reproducibility, and workflows that busy epilepsy centers can actually use (Theodore et al., 2025; Brinkmann, 2024, PMCID: PMC12702581).

There are still catches, because of course there are. Scalp EEG is noisy. HFO detection is technically finicky. Physiological fast activity can impersonate pathological signals. And this study, while impressive, is still a 42-patient cohort, not the final word. The likely future is not "goodbye invasive monitoring forever." It is more like "maybe we can walk into invasive monitoring with a much better map."

That is not a small thing. In neurology, shaving uncertainty from the process by even a centimeter can mean fewer guesses, fewer electrodes in the wrong neighborhood, and better odds that surgery targets the part of the brain actually starting the fire. Which, for a field that spends a lot of time interpreting electrical squiggles produced by a three-pound organ that routinely acts like a cryptic poet, counts as real progress.

References

1. Jiang X, Cai Z, Gonsisko C, Worrell GA, He B. Mapping epileptogenic brain using a unified spatial-temporal-spectral source imaging framework. Proc Natl Acad Sci U S A. 2025. DOI: 10.1073/pnas.2510015122. PubMed: 41359838.
2. Frauscher B, Bartolomei F, Kobayashi K, et al. High-Frequency Oscillations in Epilepsy: What Have We Learned and What Needs to be Addressed. Neurology. 2021;96(9):439-448. DOI: 10.1212/WNL.0000000000011412. PubMed: 33408149.
3. Noorlag L, van Klink NEC, Kobayashi K, Gotman J, Braun KPJ, Zijlmans M. High-frequency oscillations in scalp EEG: A systematic review of methodological choices and clinical findings. Clin Neurophysiol. 2022;137:46-58. DOI: 10.1016/j.clinph.2021.12.017. PubMed: 35272185.
4. Gonsisko CB, Cai Z, Jiang X, Duque Lopez AM, Worrell GA, He B. Electroencephalographic source imaging of spikes with concurrent high-frequency oscillations is concordant with the clinical ground truth. Epilepsia. 2024;65(12):3571-3582. DOI: 10.1111/epi.18141. PMCID: PMC11647446.
5. Theodore WH, Inati SK, Adler S, Pearl PL, McDonald CR. Imaging Epilepsy: Past, Passing, and to Come. Epilepsy Curr. 2025. DOI: 10.1177/15357597251332191. PMCID: PMC12176802.
6. Brinkmann BH. Technical Considerations in EEG Source Imaging. J Clin Neurophysiol. 2024;41(1):2-7. DOI: 10.1097/WNP.0000000000001029. PMCID: PMC12702581.

Disclaimer: The image accompanying this article is for illustrative purposes only and does not depict actual experimental results, data, or biological mechanisms.