A Brain Scan Can Tell You Which Pill to Try. It Still Can't Tell You Why You're Sad.
June 21, 2026 · Richard Passingham, Cognitive Neuroscience (A Very Short Introduction)~5 min read
A clinic's slide deck shows two brains side by side: one calm and blue, one blazing orange, captioned "your depression, visible at last." It is a persuasive picture, and it sells. But before you read the orange as the reason you feel the way you do, it's worth asking the one question the slide never answers — orange compared to what? That small question is the whole gap between a scan that genuinely helps and one that just looks like it does.
The colored blob is a difference, not a snapshot
Here is the thing almost every "brain lights up when…" headline hides. A functional MRI image is never a photograph of your brain glowing. It is the result of one condition minus another — your brain doing a task, with your brain doing a control task subtracted away. What survives the subtraction is the small remainder specific to the operation being studied. The Oxford primer Cognitive Neuroscience hammers this until it sticks: every credible image is a difference between two conditions, the way a drug trial is the drug minus the placebo. Strip away the control and you don't have a weaker result. You have no result at all, just a pretty map of a brain being a brain.
Activation is not activity, and neither is a cause
Two more quiet substitutions come with it. The orange patches are not neurons firing; fMRI measures a sluggish change in blood oxygenation, which is why the field carefully says activation, not activity. And activation, however vivid, is a readout of the brain's state during a feeling — not the reason for the feeling. The primer's sharpest example is the Pepsi challenge. Scan people who prefer Pepsi and a reward region lights up more for Pepsi. Looks like a kink in the brain. But scan Coke lovers and the same region lights up more for Coke. So the blob only ever told you which drink the person liked. It never once told you why. Now swap "drink" for "mood," and you can see the move a depression slide quietly makes.
A functional brain image is the brain's state during a feeling, not its cause — every credible scan is one condition minus a control. As a comparison (scan A vs B) it can predict which treatment is the better bet; read as a lone snapshot it merely "reveals" a cause it cannot actually see. Framework: Cognitive Neuroscience (A Very Short Introduction), Oxford. This is popular-science interpretation, not medical advice — for any health decision, consult a qualified doctor.
Why this matters right now
This isn't a museum quarrel. Through 2026 the pitch has come back dressed as precision medicine: scan plus AI plus your wearable, and an algorithm names your "biotype" and picks your drug. Some of that is real and earned. A Stanford team led by Leanne Williams scanned 801 people and sorted depression into six biotypes; for one of them, the cognitive biotype, adding the fMRI let them predict who would reach remission correctly in 63% of cases, versus 36% guessing blind (Nature Medicine, 2024). Notice the exact shape of that win, though. It is a comparison — scanned versus not-scanned, treatment that worked versus treatment that didn't. It is built from controls. That is precisely the difference logic the primer defends, and it is honest work.
The slide that crosses the line
The trouble starts when a brain image stops comparing and starts pronouncing. "This scan shows your depression is a brain disorder" is the reverse move — reading a cause out of a single colorful blob, with nothing subtracted and no one to compare you to. The primer's reply is blunt: the same pattern shows up in a perfectly healthy person who's simply been asked to dwell on sad thoughts. A divorce can carve the same groove a gene can. The scan cannot tell you which one you're looking at. So a picture that "reveals" your sadness as broken wiring is doing something the machine literally cannot do, and the orange is being asked to mean far more than it can carry. The clinical reality stays stubborn: doctors diagnose depression from your history and your symptoms, not from a snapshot.
What this means for you
You don't need a neuroscience degree to use the one tool that matters. When a scan is sold to you, ask what it was compared against, and whether it's predicting a choice or pronouncing a cause. A scan that says "people whose brains look like yours did better on drug A than drug B" is offering you a real, testable bet — take it seriously, and ask your doctor about it. A scan that says "here is your sadness, lit up" is offering you a metaphor with a price tag. Same machine, opposite honesty. The difference lives entirely in that first small question, the one the glossy slide hopes you won't think to ask.
A scan can compare two futures and say which pill is the better bet. It cannot open you up and point at the reason you ache. Sold the second when you were promised the first, you're buying a metaphor, not a diagnosis.
Before you read the orange as your verdict, ask the only question that matters: orange compared to what?
Source: framework from Cognitive Neuroscience: A Very Short Introduction (Oxford) — every brain image is a difference between conditions ("minus what?"), activation is not activity, and a blob shows a state, not its cause (the Pepsi-challenge example). Real-world basis: the 2026 wave of AI-plus-brain-scan "precision mental health" coverage (APA Monitor, Jan–Feb 2026), and the Stanford depression-biotype study (Williams et al., Nature Medicine, 2024: 801 participants; cognitive-biotype remission predicted correctly 63% with fMRI vs. 36% without). Figures per the original reports. This is popular-science interpretation, not medical advice — for any health or treatment decision, consult a qualified doctor.
一张功能性脑像是某种感受时大脑的状态、不是它的原因——每一张可信的扫描都是一个条件减去一个对照。当作比较用(扫描 A vs B),它能预测哪种治疗是更好的赌注;当作一张孤立快照读,它只是在「揭示」一个它其实看不见的原因。框架:《认知神经科学:牛津通识读本》。本文为科普解读、非医疗建议——涉及任何健康决策请咨询专业医生。
为什么这事现在尤其要紧
这不是博物馆里的旧争论。整个 2026 年,这套话术换上"精准医疗"的外衣回来了:扫描+AI+你的手环,一个算法报出你的"生物亚型",再替你挑药。其中有一部分是真本事、配得上。斯坦福莉安·威廉斯团队扫了 801 人,把抑郁分成六个生物亚型;其中一个叫"认知亚型",加上 fMRI 后,他们能在 63% 的病人身上正确预测谁会缓解,而不看脑像盲猜只有 36%(《自然·医学》,2024)。可你要看清这场胜利的精确形状:它是一次比较——扫了 vs 没扫,起效的疗法 vs 没起效的疗法。它是用对照搭起来的。这恰恰就是书里捍卫的那套差值逻辑,是诚实的活儿。
你不需要神经科学学位,就能用上那件唯一管用的工具。当一张扫描被卖给你,问它是跟什么比出来的,问它是在预测一个选择,还是在宣判一个原因。一张说"脑像跟你相似的人,用 A 药比 B 药效果更好"的扫描,是在给你一个真实、可检验的赌注——认真对待,去问你的医生。一张说"瞧,你的难过亮起来了"的扫描,是在卖给你一个带价签的比喻。同一台机器,相反的诚实。分别全在那第一个小问题里——那个亮闪闪的幻灯片,盼着你想不到要问的问题。
機能的脳画像は、ある感情のさなかの脳の状態であって原因ではない——信頼できるスキャンはすべて、一つの条件から対照を引いたものだ。比較として(スキャンA vs B)使えばどの治療がより良い賭けかを予測できるが、孤立した一枚として読めば、実際には見えない原因を「暴く」だけだ。枠組:『認知神経科学:オックスフォード・ヴェリー・ショート・イントロダクション』。本稿は科学解説であり医療助言ではない——健康上の判断は必ず専門の医師に相談を。
なぜ今これが効いてくるのか
これは博物館の中の古い論争ではない。2026年を通じて、この語り口は「精密医療」の装いで戻ってきた。スキャン+AI+あなたのウェアラブル、そしてアルゴリズムが「バイオタイプ」を名指し、薬を選ぶ。その一部は本物で、報われた仕事だ。スタンフォードのリアン・ウィリアムズらは801人をスキャンし、うつを六つのバイオタイプに分けた。そのうちの一つ「認知バイオタイプ」では、fMRI を加えることで、誰が寛解に至るかを63%の患者で正しく予測できた——脳画像を見ず当てずっぽうだと36%だ(『ネイチャー・メディシン』2024)。だがこの勝利の正確な形に注目してほしい。これは比較だ——スキャンした群 vs しない群、効いた治療 vs 効かなかった治療。対照から組み立てられている。それこそ入門書が擁護する差分の論理であり、誠実な仕事だ。