Contribution
It ran Owen et al.’s imagery test (LIT-1178) as a routine assessment of 54 patients at Cambridge and Liège, giving a first estimate of how often patients diagnosed by behaviour respond to command in the scanner: 5 of 54, 4 of 23 admitted as vegetative. It then turned the two tasks into a yes/no code, validated it on 16 healthy controls (48 of 48 answers decoded correctly) and used it with one patient, who answered five of six questions correctly while no communication could be established at his bedside.
Key insight
Two mental acts that recruit separable regions are a one-bit channel that needs no movement and no training: think of tennis for one answer and of walking through your home for the other. The answer is a judgement, a fact retrieved and assessed against the question, and the imagery is only its code, assigned by instruction. Because the questions are chosen so that the investigators do not know the answers, a correct decoding cannot be read into the data by the people reading it.
Assumptions
- Diagnosis by behaviour. Vegetative state: “repeated examinations yield no evidence of a sustained, reproducible, purposeful, or voluntary behavioral response to visual, auditory, tactile, or noxious stimuli” (p. 580). Minimally conscious state: “inconsistent but reproducible signs of awareness, including the ability to follow commands”, without interactive communication. Misdiagnosis runs at “approximately 40%” because “awareness has to be inferred from a patient’s motor responsiveness” (p. 580).
- Region specificity. Motor imagery recruits the supplementary motor area (SMA) and spatial imagery the parahippocampal gyrus (PPA). Contrasts were restricted to these two atlas regions (Harvard–Oxford), cluster z > 2.3, corrected P < 0.05 (p. 581).
- Positive findings imply awareness. The tasks need “language comprehension, working memory, decision making, or executive function”; “positive results, whether observed with or without corroborative behavioral data, do confirm that all such processes were intact and that the patient must have been aware” (p. 588). Negative findings are uninterpretable (p. 588).
- Sample. A “convenience sample” (p. 580), November 2005 to January 2009. Consent from legal guardians.
Key results
- Tasks (p. 581). Motor imagery: “imagine standing still on a tennis court and to swing an arm to ‘hit the ball’ back and forth to an imagined instructor”. Spatial imagery: “imagine navigating the streets of a familiar city or to imagine walking from room to room in their home and to visualize all that they would ‘see’ if they were there”. Two localizer scans, each five cycles of 30 s imagery and 30 s rest, cued by the spoken words “tennis”, “navigation” and “relax”.
- Responders (p. 583, Table 1). “Among the 54 patients, we identified 5 who could willfully modulate their brain activity.” All five showed SMA activation for motor against spatial imagery, four of the five PPA for spatial against motor; activity was sustained over the 30 s and locked to the cues. “Four of the five patients were considered to be in a vegetative state (including Patient 4, who has been described previously), and all five patients had a traumatic brain injury.” Table 1: Patients 4, 6, 22, 23 (admitted VS; Patient 6 motor only) and 54 (admitted MCS). “No such responses were observed in any of the patients with nontraumatic brain injuries” (p. 585).
- Bedside after the scans (p. 585). Of the four admitted as vegetative, “some behavioral indicators of awareness could be detected in two of them. However, the other two patients remained behaviorally unresponsive at the bedside, even after the functional MRI results were known and despite repeated testing by a multidisciplinary team.” The main text does not say which two.
- Communication procedure (p. 581). Before each scan a yes/no question was asked (e.g., “Do you have any brothers?”), and the participant was told to use one imagery for “yes” and the other for “no”. “The nature of the questions ensured that the investigators would not know the correct answers before judging the functional MRI data.” The cue for every answer period was the neutral word “answer”, with “relax” for rest, given once per period. One question per scan; three scans per control, six for the patient. Decoding: mean GLM estimates in the two regions, then a similarity metric (in the Supplementary Appendix, not read) comparing each communication scan with each localizer.
- Controls (p. 583). 16 healthy controls, 48 questions: “the correct answer was determined with 100% accuracy”, the answer “verified after the analysis”.
- Patient 23 (pp. 585, 588). Six autobiographical questions. Five “closely matched” one localizer “within the bounds of normal variability”, and “for those same five questions, the pattern produced always matched the factually correct answer”. Examples: “Is your father’s name Alexander?”, yes, by motor imagery; “Is your father’s name Thomas?”, no, by spatial imagery (names changed in the figure for privacy). The sixth, the last of the session, “could not be decoded … because virtually no activity was observed within the regions of interest”, and “Whether the patient fell asleep during this question, did not hear it, simply elected not to answer it, or lost consciousness cannot be determined.”
- Patient 23’s history (pp. 580–581, 585, 588). Male, 22, traumatic injury in a traffic accident. Diagnosed in a permanent vegetative state at 17 months; “confirmed by a month-long specialized assessment 3.5 years after the injury”; scanned 5 years after it (Table 1: 60.8 months), “assumed to remain in a vegetative state, although extensive behavioral testing after the functional MRI revealed reproducible, but inconsistent, responses indicative of a minimally conscious state”. “Despite the best efforts of the clinical team, it was impossible to establish any functional communication at the bedside.” The authors: “it is not known whether either ability was available during earlier evaluations”; he may have regained function, or “may have been aware during previous assessments but unable to produce the necessary motor response”; “Given that all previous assessments were based on behavioral observations alone, these two possibilities are indistinguishable” (p. 588).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | About 1 in 11 patients with disorders of consciousness in this sample (5 of 54), and 4 of 23 diagnosed vegetative, respond to imagery commands in fMRI | moderate | Table 1, p. 583; convenience sample; negatives uninterpretable |
| C2 | Responders were all traumatic; none of the non-traumatic patients responded | moderate | Table 1; small numbers |
| C3 | In some patients motor function is so impaired that bedside testing cannot reveal awareness however thorough | moderate | 2 of 4 VS responders unresponsive at bedside after fMRI was known, p. 588 |
| C4 | Two imagery tasks form a yes/no code decodable without training | strong for healthy controls, weak for patients | 48/48 controls; one patient, 5/6 |
| C5 | Patient 23 was aware and communicated | moderate | 5 correct of 6, scored blind; one patient |
| C6 | Any positive imagery response confirms awareness | weak | asserted, p. 588 |
Method
- Two localizer scans per patient (motor and spatial imagery against rest), GLM, contrasts limited to SMA and PPA.
- For communication: one question before each scan; imagery code for the answer; neutral cue “answer”.
- Decode by comparing ROI estimates on each communication scan with each localizer, by a similarity metric (Supplementary Appendix).
Concepts
- Willful modulation. “Voluntary, reliable, and repeatable blood-oxygenation-level–dependent responses in predefined neuroanatomical regions when prompted to perform imagery tasks” (p. 585).
- Functional communication. Here, correct yes/no answers to questions whose answers the investigators did not know.
Connections
It extends Owen et al. 2006 (LIT-1178, read in NOTE-962), whose patient is Patient 4 here, and Boly et al. 2007 on the imagery paradigm. It cites Weiskopf et al. on real-time fMRI brain–computer interfaces, but does not use real-time feedback. Its diagnostic categories are Giacino et al.’s (2002) minimally conscious state and the Multi-Society Task Force’s vegetative state.
Bearing on the record
- CASE-107: the CASE and the LIT run two patients together. The CASE says “5 of 54 patients … could modulate their brain activity at will, two of them with no voluntary behaviour detectable at the bedside, and the imagery could be used to answer yes or no”. The LIT’s “Why this is here” says “answers given, through mental imagery by patients with no detectable voluntary behaviour”. The paper does not support that reading. Only one patient was asked questions, Patient 23, and on retesting he showed “reproducible, but inconsistent, responses indicative of a minimally conscious state” (p. 581). The two patients with no behaviour at the bedside were not asked questions. So the paper shows, separately, (a) awareness-on-command with no detectable behaviour, in two patients, and (b) yes/no judgement through imagery in a patient who still had inconsistent motor responses, though no bedside communication. Neither patient shows both.
- CLAIM-341, defeater (b). What the paper shows is judgement given with no action on the world in the channel that carried it: the answer went out through imagery read by a scanner. It does not show judgement in a patient with every contingency between intention and outcome removed. Patient 23 kept inconsistent voluntary motor responses, and the two unresponsive patients were only undetected, not shown to have no motor channel. The authors themselves call bedside unresponsiveness a limit of detection: “motor function can be so impaired that bedside assessments … may not reveal awareness” (p. 588). So it is evidence for the thesis’s weaker reading (judgement runs, here, without needing action) and not a test of its defeater.
- What the judgement was about. The questions shown are autobiographical facts from before the injury: his father’s name. They test retrieval and assessment of an old model, built while acting, not whether a model of the present is kept in touch five years into near-total immobility. That bears on the CASE’s “What it cannot show”. The other four questions are in the Supplementary Appendix, not read.
- CLAIM-337: the channel is imagined action. The “yes” code was to “imagine standing still on a tennis court and to swing an arm” and is read from the supplementary motor area. Like Owen et al., this is motor imagery read from a motor region. The judgement and the channel separate cleanly here, which the thesis can use: the judgement (is the name right?) is not itself an imagined act; it is encoded into one by an arbitrary convention set by the investigators. The counter can still say the encoding is mental action. The thesis’s answer, that it acts on the patient’s own state and not the world, holds.
- The asymmetry. The paper exhibits it twice. Negative findings are uninterpretable in 49 of 54 patients (p. 588). The sixth question, which produced no signal, cannot be told apart from sleep, deafness, refusal or loss of consciousness (p. 585). Evidence of lucidity arrives only when the patient drives the channel.
- CLAIM-342, from the observers’ side. The design is an instance of the claim’s anchor, but for the investigators’ interpretation, not the patient’s: “The nature of the questions ensured that the investigators would not know the correct answers before judging the functional MRI data” (p. 581). That is scoring not routed through the interpretation it tests (that the patient is aware), and it is what makes the 5-of-6 more than a reading of noise. For the patient’s own loop the paper says nothing: no feedback was given, so no contingency between his intention and an outcome was restored in the session. Against Chaudhary et al. (LIT-1200), this is communication with no neurofeedback and no training.
- QUESTION-121. Nothing directly.
- The LIT summary. Correct in numbers. It should say the yes/no procedure was used with one patient, who was not one of the two without bedside behaviour, and give the 5-of-6 and 48-of-48 figures.
- For the anthology. No instruction for machine-learning practice.
Limitations
- One communicating patient, six questions, one undecodable.
- Convenience sample; negatives uninterpretable; detection depends on traumatic aetiology in this sample, unexplained.
- The similarity metric and the full question list are in the Supplementary Appendix, not read here.
- The authors’ “must have been aware” for any positive result is asserted.
- The main text does not identify the two behaviourally unresponsive responders or give their histories.
Open questions
- Whether the two behaviourally unresponsive responders could answer questions. That would put judgement and the absence of detectable behaviour in one patient.
- Whether a patient’s answers about the present (where they are, what happened today) are as accurate as about the past. That would bear on whether the model is kept in touch, which is CLAIM-341’s question, and not only retained.