AI-Guided TMS Targeting: How Machine Learning Is Making Brain Stimulation More Precise

Everything you need to know about AI-Guided TMS Targeting: How Machine Learning Is Making Brain Stimulation More Precise — how it works, what it costs, and how to find a provider who actually knows what they're doing.

TMS therapy has helped hundreds of thousands of people with treatment-resistant depression. But here is something most people do not hear during their consultation: the standard method for aiming the magnetic coil has not changed much since the 1990s.

Related: Stanford Neuromodulation Therapy (SNT) · The History of TMS: From Faraday to FDA and Beyond · Accelerated TMS Protocols

The most common approach is the 5-cm rule. Measure 5 centimeters forward from the motor cortex, place the coil there, and call it the dorsolateral prefrontal cortex. It is simple, requires no imaging, and it is the method used in the pivotal trials behind TMS’s FDA clearance for depression. But well enough leaves room to do better.

Brain anatomy varies significantly between individuals. Your dorsolateral prefrontal cortex is not in exactly the same spot as mine, and the specific subregion that connects to mood-regulating circuits varies from person to person, and the TMS coil’s focal point is small.

Miss the sweet spot and you are stimulating brain tissue next to, but not part of, the circuit you need. The treatment might still work partially. Or it might not work at all. Researchers have suggested that targeting variability may be one reason many patients do not respond; real-world studies report response rates of roughly 50-60% (Carpenter et al., 2012).

Artificial intelligence is finally cracking this problem open.

What You’ll Learn

• Why inconsistent targeting has held TMS outcomes back
• How AI-guided targeting works using neuroimaging and machine learning
• What outcome improvements AI targeting actually delivers
• What imaging costs and where to get it
• The technology landscape and what it means for you right now

How AI-Guided Targeting Works

The basic idea is intuitive, even if the engineering behind it is genuinely complex. You get a brain scan, usually a functional MRI, sometimes combined with structural MRI and diffusion tensor imaging. Some research groups feed such scans to machine learning algorithms trained on previous scans from people whose TMS outcomes are known.

The AI does two things. First, it maps your individual brain connectivity, pinpointing the exact cortical target with the strongest functional connection to deeper mood-regulating structures like the subgenual anterior cingulate cortex. That is the where to aim problem.

Second, and this is newer, some systems predict how likely you are to respond to TMS at all, based on your brain connectivity patterns and clinical features. That is the should we aim problem. Equally important, but historically ignored.

Several commercial platforms have emerged to make this practical. Some companies and academic groups have built software that takes neuroimaging data and produces a specific coil placement coordinate, often rendered as a 3D target that clinicians can see during setup.

From your perspective, it is straightforward. You get a brain scan, typically under an hour in the MRI machine, before your TMS course begins. The scan gets uploaded, the software processes it, and your clinician gets a personalized targeting map. Your TMS sessions then use neuronavigation, a GPS-like system that tracks the coil’s position relative to your brain anatomy in real time, to hit that target consistently.

The Outcome Improvements Are Real

The evidence is still developing. Some studies have compared fMRI-guided targeting to standard anatomical targeting, but results are mixed and the field has not settled on how much targeting adds on its own.

The best-known example is Stanford’s work: fMRI-guided accelerated theta burst stimulation produced a larger drop in depression scores than sham in a small randomized trial (29 participants), but it is difficult to say how much came from the targeting and how much from the accelerated dosing.

The Stanford Neuromodulation Therapy protocol combines fMRI targeting with accelerated stimulation. Its open-label study reported remission in about 90% of 22 participants (Cole et al., 2020), while the sham-controlled trial reported a mean 52.5% reduction in depression scores with active treatment versus 11.1% with sham (Cole et al., 2022). Both targeting and dosing likely matter.

Prediction is an active research area too. Research models using baseline brain connectivity and clinical variables have shown some ability to predict TMS response, but these tools are not validated for routine clinical decisions.

If reliable prediction becomes available, it could help steer people predicted to be non-responders toward different coil placements, different stimulation parameters, or other treatments.

What Imaging Costs and Where to Get It

The added cost of neuroimaging is the main barrier right now. The cost of the imaging varies widely by location, imaging center, and whether insurance covers it. Some centers bundle the scan into their overall TMS treatment package. Others bill it separately.

Insurance coverage for the neuroimaging piece is inconsistent. Coverage for MRI used for TMS targeting is often not guaranteed, and some insurers question the medical necessity of fMRI specifically for TMS targeting. Prior authorization and a letter of medical necessity from your provider can help. Check our insurance guide for carrier-specific information.

Not every TMS clinic offers AI-guided targeting yet. The ones that do tend to be academic medical centers, larger specialty practices, and clinics that have invested in neuronavigation equipment. You can search our directory to find providers that advertise precision targeting or neuroimaging-guided TMS.

If the clinic nearest to you does not offer fMRI-guided targeting, some providers work with remote imaging analysis services. You get the MRI locally, the data gets sent to a specialized center for processing, and the targeting coordinates come back within a few days.

The Technology Landscape in 2026

Several distinct approaches are competing and, increasingly, combining.

Functional connectivity targeting uses resting-state fMRI to map individual brain networks. This is the approach behind the Stanford SNT protocol and has the most clinical evidence of these approaches, though it is still limited. It identifies the best stimulation site based on how your prefrontal cortex connects to subcortical mood circuits.

Structural targeting with AI optimization uses high-resolution structural MRI plus machine learning to predict the best target without requiring functional imaging. Potentially cheaper and faster, though possibly less precise. This is largely experimental.

EEG-guided targeting is the budget option. Electroencephalography is cheap, portable, and fast. Researchers are exploring whether EEG can help estimate optimal targets. This is largely experimental and has not been shown to outperform standard targeting in routine practice.

Real-time adaptive protocols are the frontier. These systems adjust stimulation parameters during treatment based on neural responses measured via EEG or fNIRS. Think of it as closed-loop TMS. The machine monitors your brain’s reaction and tweaks the protocol on the fly. Early results are promising, but this is still largely experimental.

The broader direction is clear. TMS is moving from a one-size-fits-all treatment to a precision medicine approach where the protocol is tailored to your neurobiology.

What This Means for You Right Now

If you are considering TMS or have tried it before without enough results, AI-guided targeting is worth asking about.

If you are a first-time TMS patient and you have access to a clinic that offers neuroimaging-guided targeting, the added cost of the brain scan may or may not improve your odds of response; the evidence is still developing. Whether it is worth it depends on your finances, insurance, and your clinician’s advice.

If you tried standard TMS and it did not work, poor targeting is a plausible explanation. Before concluding that TMS does not work for me, consider a course with precision targeting. Some standard-TMS non-responders have responded when re-treated with different protocols, though how often is not well established.

If you are choosing between providers, ask about their targeting method. A clinic using neuronavigation and neuroimaging is working with more individualized information than one using the 5-cm rule, though better outcomes are not guaranteed. Our specialist directory can help you find providers who use advanced targeting approaches.

Five years from now, treating everyone with the same coil placement will likely seem as outdated as prescribing the same antidepressant dose to every person regardless of weight, genetics, or metabolism. The tools for personalization exist today. They are just not standard yet.

Related reading

• TMS for Multiple Sclerosis: Addressing Fatigue, Depression, and Cognitive Impairment
• What TMS Actually Feels Like: A Session-by-Session Guide
• How to Prepare for Your First TMS Session
• Preparing for Your First TMS Session: A Practical Walkthrough
• TMS for Treatment-Resistant Depression: What Patients Need to Know