Daniel sent us this one — he points out that fMRI was one of those genuinely decisive shifts in psychiatry, the moment where clinical syndromes that had only been describable by patients suddenly became visible on a scan. And that mattered enormously for patients who'd been dismissed as having conversion disorders. But fMRI is no longer the cutting edge. The question is, what comes next in imaging, what understanding might that unlock, and — the part I think is actually the sharpest — will any of these advanced tools ever actually reach day-to-day clinical practice, or are they stuck in research labs forever?
This is exactly the right moment to ask this, because we're in this strange limbo. fMRI has been around long enough that everyone's heard of it, it sounds futuristic, but the actual frontier has moved well past it. And that gap between what exists in research and what a psychiatrist can actually order on a Tuesday afternoon is enormous.
Let's start with what fMRI actually gave us. Because the prompt's right — it was a genuine inflection point. What did we learn that we literally couldn't have known otherwise?
The big one, and I still think this is underappreciated, is the default mode network. Before fMRI, nobody had a coherent picture of what the brain does at rest. You'd assume it just powers down. But fMRI showed this interconnected set of regions — medial prefrontal cortex, posterior cingulate, angular gyrus — that light up when you're doing nothing in particular. And it turns out that network is hyperactive in depression, in rumination, in anxiety. The brain at rest isn't resting — it's narrating, it's worrying, it's rehearsing social failures. That was invisible before.
The brain's internal podcast, and the host won't stop talking.
That's not a bad way to put it. And then you get the connectivity studies. Not just "this region activates during sadness" but "in depression, the amygdala and the prefrontal cortex stop talking to each other properly." The regulatory circuit breaks down. You can see the disconnection. That gave us a mechanistic model of depression that wasn't just neurotransmitter soup — it was a wiring problem, a communication problem.
Which also explains why the serotonin story was always too simple. If it were just a chemical bath, you wouldn't see circuit-level dysregulation.
And fMRI helped kill the chemical imbalance myth, or at least wound it badly. But here's what I find fascinating — fMRI also gave us predictive biomarkers. There was a 2013 study by Helen Mayberg's group where they scanned depressed patients before treatment, and the resting-state activity in the anterior insula predicted whether they'd respond to CBT or to an SSRI. That's useful. If you have low insula activity, medication might be the better bet. If it's high, therapy might work better. That's personalized psychiatry, and it came directly from fMRI.
Yet I can't walk into a clinic and get that scan. That's the tension at the heart of the prompt, isn't it? We have the knowledge, but not the access.
That's exactly where we are. And to answer the first part of the question — what comes next — we have to look at what fMRI can't do. The spatial resolution is decent, a few millimeters, but the temporal resolution is terrible. The blood oxygen signal lags behind neural activity by seconds. A lot happens in a brain in two seconds. And the signal is indirect — you're not seeing neurons fire, you're seeing the plumbing respond to the electrical bill.
What replaces it?
A few things, and they're at different stages of maturity. The one I'm most excited about is portable, wearable functional near-infrared spectroscopy — fNIRS. It measures the same hemodynamic signal as fMRI but uses light instead of giant magnets. You wear a cap with optodes, and you can scan someone while they're walking, talking, interacting socially. No tunnel, no noise, no lying perfectly still.
The scanner as a baseball cap.
And the applications for psychiatry are huge. Imagine doing exposure therapy with someone who has social anxiety, and you can watch their prefrontal cortex in real time as they navigate a conversation. You can give them neurofeedback — "your anxiety circuit is activating, let's practice downregulating it right now." That's a completely different therapeutic model.
This exists now?
It exists in research labs. The resolution isn't as good as fMRI, and the signal gets noisy with hair, with motion, with skull thickness. But it's improving fast. There was a 2024 review in NeuroImage that showed fNIRS-based neurofeedback for ADHD producing effect sizes comparable to medication in some studies. That's not small.
A research-grade fNIRS system might run you fifty to a hundred thousand dollars. That sounds like a lot, but an fMRI machine is one to three million, plus the helium, plus the shielded room, plus the technician. The fNIRS cap is the laptop of brain imaging.
Which makes it plausible for clinics. That's the first thing we've mentioned that might actually answer the access question.
But let me give you the next step beyond that. There's a technique called optically pumped magnetometry — OPM-MEG. Traditional MEG requires a giant, liquid-helium-cooled scanner and a magnetically shielded room. OPM-MEG uses these tiny vapor cell sensors that work at room temperature, and they can be mounted on a helmet. You get the temporal precision of MEG — milliseconds — with the portability of fNIRS. You can actually watch neural oscillations in real time as someone moves around.
I've seen photos of those helmets. They look like something from a cyberpunk costume department.
And they're still mostly in the UK — Nottingham, UCL — but the technology is moving toward commercialization. The company Cerca Magnetics spun out of Nottingham in 2021, and their systems are being bought by research hospitals now. Give it five to ten years.
Again, we're talking about hardware. The prompt asks what greater understanding might come from these advances. What's the big conceptual leap?
I think the leap is from static mapping to dynamic circuit tracking. fMRI gave us snapshots — the depressed brain at rest, the anxious brain during threat processing. The next generation gives us movies. You can watch a mood state emerge in real time. You can see the cascade — a triggering thought, amygdala activation, prefrontal recruitment or failure, the behavioral output. Psychiatry has never had access to that causal chain in living humans. We've inferred it from animal models and post-mortem tissue and patient report. Now we can watch it unfold.
We move from "this region is implicated in depression" to "this is the sequence of events that produces a depressive episode.
And that changes treatment. If you know the sequence, you know where to intervene. Is the problem that the amygdala is overreacting, or that the prefrontal cortex is failing to regulate a normal reaction? The treatment is different. One calls for dampening the initial response, the other calls for strengthening top-down control. fMRI gave us hints of this. The new tools will let us see it clearly enough to guide decisions.
What about the molecular level? All of these are still looking at blood flow or magnetic fields. They're not seeing receptors, not seeing neurotransmitter dynamics.
That's where PET comes back in. PET never went away, but it's been limited by the need for radioactive tracers and the terrible spatial resolution. What's changed is the development of new tracers. There's now a tracer for the kappa opioid receptor, which is deeply involved in stress and dysphoria. There are tracers for neuroinflammation — translocator protein, or TSPO, markers. There's a tracer for synaptic density itself, called UCB-J, that binds to SV2A, a protein found in basically every synapse. You can literally count synapses in a living human brain.
That's not imaging — that's an inventory.
It's clinically relevant. The Yale PET Center published work in 2023 showing reduced synaptic density in depression, in PTSD, in schizophrenia. These aren't just functional problems — there's actual loss of connection points. And we can watch that change with treatment. That's a completely different way of thinking about what psychiatric illness is.
You've got the circuit-level tools — fNIRS, OPM-MEG — giving you the dynamic movie, and the molecular tools — new PET tracers — giving you the synaptic inventory. Do they talk to each other?
That's the holy grail. Simultaneous PET-fMRI already exists in a few research centers. You get the molecular specificity and the functional connectivity at the same time. But the machines are monstrously expensive and the radiation dose limits repeat scanning. The future is probably hybrid systems that combine the wearable MEG or fNIRS with some form of molecular imaging, or with blood biomarkers that give you the chemical readout while the scanner gives you the circuit readout.
Let me pull us back to something the prompt raised that I think gets skipped in most coverage. fMRI helped patients who were dismissed as having conversion disorders — people whose symptoms were written off as not real because you couldn't see them. What's the equivalent for these new tools? Who gets validated next?
I'd point to two groups. The first is people with so-called "medically unexplained symptoms" — chronic fatigue, fibromyalgia, certain pain syndromes. Neuroimaging is starting to show real, replicable abnormalities in pain processing circuits, in glial activation, in connectivity between the insula and the default mode network. These patients have been told for decades that it's in their heads. Well, it is in their heads — in the sense that their brain is doing something measurably different.
The distinction between "in your head" and "in your brain" doing a lot of heavy lifting there.
It shouldn't have to. The brain is an organ. If it's malfunctioning, that's a medical problem, full stop. The second group is people with personality disorders, particularly borderline personality disorder. There's been a long history of stigma — these patients are "difficult," they're "manipulative." fMRI and the newer tools have shown clear differences in emotion regulation circuits, in the amygdala-prefrontal connectivity we talked about. It's not a character flaw — it's a circuit disorder. That reframes the entire clinical conversation.
It also reframes the legal conversation, which nobody wants to have. If we can see these things on a scan, what does that do to criminal responsibility? To how we think about free will?
We're already seeing this in the courts. fMRI evidence has been introduced in sentencing, usually as mitigation — "look at this defendant's prefrontal cortex, it's underdeveloped, they couldn't control their impulses." The problem is that fMRI is a group-level tool. It tells you about populations, not individuals. You can say "people with this pattern have impaired impulse control" but you can't say "this specific person couldn't control their impulse at the moment of the crime." The newer tools with better temporal resolution might actually make that worse — you'll get a more detailed picture that invites more overinterpretation.
The CSI effect, but for psychiatry.
The brain scan looks scientific, looks definitive, and juries are going to treat it as ground truth even when the science doesn't support that level of inference. We're going to need really careful forensic standards.
Let's go back to the access question, because I think that's where the prompt lands hardest. We've described a future where we have portable brain scanners, synaptic counting, real-time circuit monitoring. Is any of this going to be available to the person walking into a community mental health clinic in Tulsa or Tel Aviv?
The honest answer is that it's going to be uneven, and it's going to be slow. But there are reasons for optimism that didn't exist ten years ago. fNIRS is the most promising for broad deployment. The hardware is getting cheaper, the analysis pipelines are getting automated, and you don't need a physicist to run it. I could imagine a future where every psychiatry residency teaches fNIRS interpretation the way cardiology teaches EKG reading. It becomes a standard tool, not a research luxury.
What's the barrier? Is it cost, training, or evidence?
All three, but evidence is the big one. For a tool to enter clinical practice, you need to show that it changes outcomes — not just that it produces interesting pictures. You need randomized controlled trials where patients randomized to imaging-guided treatment do better than patients randomized to treatment as usual. Those trials are expensive and slow, and the imaging companies aren't going to fund them the way pharma funds drug trials because the profit margin on a scanner cap is much lower than on a patented molecule.
There's a market failure. The thing that would improve care doesn't get built because nobody gets rich enough from building it.
That's part of it. But there's also a cultural problem in psychiatry. The field has been through a century of biological promises that didn't pan out. The monoamine hypothesis, the candidate gene era, the "decade of the brain" that didn't transform clinical care. There's skepticism, and it's earned. Every new imaging modality has to prove it's not just another pretty picture.
Speaking of pretty pictures — and this is a genuine question — how much of the fMRI literature actually replicates? Because I've seen the dead salmon study.
Right, the dead salmon. For listeners who haven't heard this — in 2009, a group put a dead Atlantic salmon in an MRI scanner, showed it photos of humans in social situations, and ran standard statistical analysis. They found significant activation in the salmon's brain. The point wasn't that fMRI is fake — it was that if you don't correct for multiple comparisons properly, you'll find signal in noise. The salmon became a cautionary tale.
The salmon's posthumous social cognition. A remarkable finding.
Truly ahead of its time. But to answer your question seriously — the replication situation has improved dramatically since the salmon days. The field adopted stricter correction methods, preregistration became more common, the Human Connectome Project set standards for acquisition and analysis. But there's still a file drawer problem. And the sample sizes in a lot of fMRI studies are tiny — twenty people, thirty people. You can't build clinical tools on n equals twenty.
Which is why the big consortium studies matter.
The ABCD Study — Adolescent Brain Cognitive Development — has over eleven thousand kids, longitudinal, with fMRI at multiple time points. The UK Biobank has scanned over fifty thousand adults. When you have those numbers, you can start to see patterns that are reliable enough for clinical use. But those studies are publicly funded, and they take decades.
Let me ask something the prompt didn't ask directly but I think is implied. We've been talking about psychiatry, but a lot of these tools were developed for neurology — for stroke, for tumor localization, for epilepsy surgery planning. Is psychiatry actually driving the next generation, or is it riding neurology's coattails?
It's mostly riding coattails, honestly. The big investments in OPM-MEG are for epilepsy focus localization. The PET tracer development is driven by Alzheimer's research — amyloid and tau imaging. Psychiatry benefits from the spillover, but it's not the primary customer. And that matters because the clinical questions are different. A neurologist wants to know "where is the seizure starting." A psychiatrist wants to know "will this person respond to this antidepressant." The tools need to be validated for different endpoints.
Psychiatry needs its own validation pipeline, not just borrowed hardware.
And that pipeline is starting to emerge. The field has a name for it — computational psychiatry. It's the intersection of computational neuroscience, machine learning, and clinical psychiatry. The idea is that you don't just look at a brain scan and say "the amygdala is active." You build a mathematical model of the patient's decision-making, their learning processes, their reward prediction errors, and you fit that model to their behavioral and neural data. Then you can say "this patient has a specific deficit in reward learning, not in reward sensitivity, and here's what that means for treatment.
Instead of "your amygdala is overactive," it's "your prediction error signal is miscalibrated in a specific way that explains why you can't experience pleasure.
That's the ambition. And it's testable. You can run someone through a reinforcement learning task in a scanner, fit the model, and generate a patient-specific parameter that predicts treatment response. There was a 2022 study in Nature Medicine from the Max Planck group where they did exactly this for depression — used a computational model of effort-based decision-making to predict who would respond to a dopamine-acting drug versus a serotonin-acting drug. That's precision psychiatry.
The model runs on fMRI data, so we're back to the access problem. But presumably these models could run on fNIRS or EEG data too.
EEG is actually the sleeper here. Everyone forgets about EEG because it's old, it's low-resolution, it's the technology your grandfather's neurologist used. But EEG has millisecond temporal resolution, it's cheap, it's portable, and the computational methods for source localization have gotten dramatically better. You can't see the amygdala directly with EEG, but you can infer its activity from the electrical patterns at the scalp. Combine high-density EEG with machine learning classifiers, and you can start to do a lot of what fMRI does, at a fraction of the cost, in a regular exam room.
The future might not be a fancy helmet. It might be a shower cap full of electrodes and a laptop.
That's the most likely near-term clinical scenario. And it's already happening in some places. There are companies — I'm thinking of a couple in the Boston area and one in Tel Aviv — that are building EEG-based diagnostic aids for ADHD, for depression subtypes, for early cognitive decline. They're not replacing the clinician, they're giving the clinician a quantitative readout to supplement the interview. That's the realistic first step.
The prompt asks whether we'll ever get to the point where these tools are available to the general public, not just research. It sounds like your answer is yes, but it's going to be EEG and fNIRS, not the million-dollar magnets.
Yes, with an important caveat. "Available to the general public" could mean two different things. It could mean "available to any psychiatrist in clinical practice," which I think is ten to fifteen years away for EEG-based tools and maybe fifteen to twenty for fNIRS. Or it could mean "available directly to consumers," like a brain scan you order on your phone. That's a different question entirely.
Which is already happening, sort of. There are direct-to-consumer EEG headsets. There are companies that will sell you a "brain health assessment" based on a scan.
Most of it is pseudoscience dressed up in nice graphics. The data quality is terrible, the normative databases are nonexistent, and the clinical claims are unsupported. But the demand is real. People want to see their brains. They want the validation that comes from an image. That's not going away.
It's the same impulse that makes people share their genetic ancestry results. "This is who I am, scientifically certified.
Just as likely to be misinterpreted. The difference is that a brain scan feels more immediate than a genotype. You can look at it and think you see your depression, your anxiety, your trauma. The risk of reification — of treating the image as the illness itself — is very high.
Which brings us back to the conversion disorder point. The whole reason fMRI was a step forward for those patients is that it made the invisible visible. But the risk of making things visible is that you start to believe only what you can see.
That's the central tension in biological psychiatry. We want to validate suffering by finding its biological signature. But if we tie diagnosis too tightly to biomarkers, we risk excluding people whose suffering is real but doesn't show up on our particular scan. The scan becomes the gatekeeper. And scans are imperfect. They have false negatives. They have technical artifacts. They depend on the patient being able to lie still, follow instructions, not have too much head motion. All of those introduce bias.
The future of imaging is also a future of new forms of exclusion, unless we're very careful about it.
And I think the way to be careful is to treat these tools as decision support, not as oracles. The psychiatrist still does the interview. The scan adds a data point. The combination is better than either alone. That's the model that actually works, and it's the model that's most likely to be adopted in real clinical practice.
Let me ask about something we haven't touched. The prompt mentions fMRI as a recent example of an advance that unlocked new knowledge through research trials. What's the equivalent for the next generation? What question can we only answer with these new tools?
I think the biggest open question is how psychotherapy changes the brain in real time. We know from before-and-after fMRI scans that CBT changes prefrontal-amygdala connectivity. But we don't know what happens during the session itself. What's the neural signature of a therapeutic insight? Of a corrective emotional experience? Of the moment when a patient reframes a traumatic memory? Those are the active ingredients of therapy, and we've never seen them directly.
Because you can't do therapy inside an MRI tube.
You can't. But you can do therapy wearing an fNIRS cap. You can do it with EEG. Suddenly the black box of the therapeutic process becomes observable. And that could change how we train therapists, how we tailor interventions, how we understand what actually works. Not just "CBT is effective for depression" but "in this specific moment, this specific intervention produced this specific neural change that predicted this specific clinical improvement." That's a completely different level of understanding.
A completely different level of accountability for therapists.
Which is going to be uncomfortable for the field. But ultimately good for patients.
What about the question we haven't asked — what if none of this pans out? What if the brain is just too noisy, too individual, too context-dependent for imaging to ever be clinically useful in psychiatry?
That's a real possibility, and I think it's under-discussed. The brain is not like the heart. Cardiac function is relatively straightforward — it pumps blood. Brain function is entangled with meaning, with context, with the specific content of thoughts. The same amygdala activation could mean fear, or excitement, or novelty detection, depending on what else is happening. The signal is inherently ambiguous. It's possible that no imaging modality will ever give us the kind of diagnostic certainty we get from a cardiac stress test or a tumor biopsy.
The brain as the organ that refuses to be reduced to an organ.
And that's not a mystical claim — it's a claim about complexity. The number of possible brain states vastly exceeds the number of measurements we can take. We're always going to be undersampling. The question is whether we can sample enough to be clinically useful, even if we can't sample enough to be definitive.
I think that's a good place to land the core discussion. The prompt asked three things: what comes next, what understanding might it unlock, and will it ever reach day-to-day medicine. We've got portable circuit-level tools, molecular synaptic counting, computational models that read out specific decision-making deficits, and a realistic path to clinical deployment that probably runs through EEG and fNIRS rather than giant magnets. The understanding unlocked is the shift from static maps to dynamic causal sequences, and the validation of patient groups who've been dismissed. And the access question — yes, but unevenly, and with real risks of new exclusions and overinterpretation.
That's a fair summary. The one thing I'd add is that the timeline matters. These transitions take longer than anyone expects. fMRI was invented in 1990, and it's still not a routine clinical tool in psychiatry thirty-six years later. The next generation will probably take as long, for the same reasons — evidence, cost, training, culture. The technology will be ready before the system is.
Now: Hilbert's daily fun fact.
Hilbert: The P2 tiling, a specific semi-regular tessellation of the plane involving equilateral triangles and regular hexagons, was long attributed to the New Zealand mathematician Alexander Aitken, who sketched it in a 1934 letter from Dunedin. It wasn't until 1987 that a review of Aitken's correspondence revealed he had copied the pattern from an unpublished notebook of the English schoolteacher and amateur geometer Marjorie Rice, who had discovered it four years earlier while experimenting with pentagonal tilings on her kitchen table.
Kitchen table geometry corrections. That's a niche.
The interwar tiling attribution drama I didn't know I needed.
This has been My Weird Prompts, produced by Hilbert Flumingtop. You can find every episode at myweirdprompts.If you want more on the computational psychiatry angle, the Max Planck group's 2022 paper in Nature Medicine is worth your time — we'll link it in the show notes. I'm Corn.
I'm Herman Poppleberry. See you next time.
