Update: Serious methodological concerns have been raised about the Debattista paper, check out Neuroskeptic’s coverage here.
In 1996 we witnessed a computer beat the world chess champion, something many never thought would be possible. Is it possible that in 2011 a computer could actually beat psychiatrists in something as intrinsically human as diagnosing mental disorder and even deciding the most effective medication? According to preliminary results from research currently being conducted by a team at Stanford University this is already rapidly becoming a reality.The key question for psychiatrists today is not the naming of the disorder but deciding (when a prescription is necessary) which prescription is most likely to be effective. The wrong prescription can do much more harm than good. It is accepted that to a large extent psychiatrists still rely on trial and error. Unlike other areas of medicine, psychiatric problems tend to lack biomarkers that inform pharmacotherapy such as bacterial assays that guide antibiotic treatment or histological and genetic tests that guide chemotherapy. The massive STAR*D clinical trial of antidepressants demonstrated just how much of a lottery the choice can be.
A study published in the January volume of the Journal of Psychiatric Research by a team at Stanford demonstrates a computer system that appears to be able to tackle this problem in the precribing of anti-depressant medicaton . The method used is called “referenced EEG” (rEEG). This uses mathematical algorithms to compare the brain patterns of a patient to a database of the brain patterns of previous patients with a similar condition and their treatment outcomes. Essentially the patient is given the treatment that is demonstrated to work best for patients with similar brain patterns.
This technique has been suggested before but only now is it seriously beginning to present a major challenge to the traditional method. In November a research group in Canada demonstrated an rEEG method that categorised depressive, bipolar and schizophrenia patients with 85% accuracy. A month later the same researchers published another paper demonstrating that the program successfully classified whether Schizophrenia patients would respond positively to clozapine, again in 85% of cases. Now the Stanford team led by Charles DeBattista has published preliminary findings that appear to demonstrate rEEG can select notoriously hard to predict depression medications with 65% accuracy. This is significantly higher than the 38% score achieved using the STAR*D approach which is widely considered best practice amongst Psychatrists.
The critically minded amongst you may well baulk at the methodological conundrums involved in comparing an rEEG diagnosis to a human one. If these results are valid, they are truly astounding. In 1949 Ash demonstrated that only 20% of Psychiatrists agreed on diagnosis, as recently as 1962 that figure was only 42%. More recently the “DSM” has assured agreement is now closer to 90%. Whether the DSM diagnosis is valid is another debate however. The suggestion that rEEG may be able choose an appropriate prescription after a human psychiatrist has performed a diagnosis certainly seems more tangible a possibility at this point in time.
It is important to recognise the findings are only preliminary. There are always methodological issues inherent in a pilot study that prevent results being as earth shattering as they may sound. The medications prescribed by the rEEG program were far more varied and qualitatively different from the limited selection of STAR*D. The issue may be that psychiatrists are exercising greater restraint in prescribing higher risk medication at the expense of better results. (This is in no way a criticism of psychiatrists, caution is obviously of paramount importance when dealing with such powerful medications.) Regardless, research groups around the world are joining the race to test and expand the method. Psychiatrists (and EEG technicians) will doubtless be awaiting these results with bated breath.
If this news came as a shock to you I’d recommend taking a look at the work of Ray Kurzweil, a remarkable professor who has made some amazing discoveries himself and continues to make astounding technological predictions.
Khodayari-Rostamabad A, Reilly JP, Hasey G, Debruin H, & Maccrimmon D (2010). Diagnosis of psychiatric disorders using EEG data and employing a statistical decision model. Conference proceedings : … Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Conference, 2010, 4006-9 PMID: 21097280
Khodayari-Rostamabad, A., Hasey, G., MacCrimmon, D., Reilly, J., & Bruin, H. (2010). A pilot study to determine whether machine learning methodologies using pre-treatment electroencephalography can predict the symptomatic response to clozapine therapy. Clinical Neurophysiology, 121 (12), 1998-2006 DOI: 10.1016/j.clinph.2010.05.009
Charles DeBattista, Gustavo Kinrys, Daniel Hoffman, Corey Goldstein, John Zajecka, James Kocsis, Martin Teicher, Steven Potkin, Adrian Preda, Gurmeet Multani, Len Brandt, Mark Schiller, Dan Iosifescu, Maurizio Fava (2011). The use of referenced-EEG (rEEG) in assisting medication selection for the treatment of depression. Psychiatric Research, 15 (12), 64-75 DOI: The use of referenced-EEG (rEEG) in assisting medication selection for the treatment of depression
Continue reading »Follow Simon on Twitter, Facebook, Google+, RSS, or join the mailing list.
So, theres been a lot of chatter over recent weeks about Brain Computer Interfacing since Tan Le’s TED talk on the Emotiv which resulted in the emotiv site being overloaded with hits and the letters “B, C and I” becoming familiar to fellow geeks the world over. The talk is mostly a marketing pitch so as I’m currently (amateur that I may be) involved in some BCI research myself I figured I’d give a little explanation of whats going on here…
Brain Computer Interfacing is not a new discovery but is currently emerging as a practical option for clinical treatment of “locked in syndrome”. The claims made by Emotiv however that the company has made a discovery that allows BCI to be released on the consumer market for gaming purposes are therefore intriguing. A major problem for decoding the EEG signals that the BCI relies upon is that the electrical signal the brain gives off is very weak and must travel through both skull and tissue. Emotiv claims it’s “breakthrough was to create an algorithm that unfolds the cortex so that we can map the signals closer to it’s source and therefore making it capable of working across a mass population“.
This currently hasn’t really been achieved by the army of BCI researchers working in institutions around the world so if true it is a genuine breakthrough that could have positive implications for sufferers of many degenerate brain disease patients and paralysis patients as well of course as the shock and awe of gamers everywhere. This finding however hasn’t been published in any journals to date, this should ring a number of alarm bells.
The second eye brow raiser is that even relatively old EEG equipment costs tens of thousands of dollars and requires close contact between fine metal electrodes which must be carefully placed exactly on top of very precise areas of the scalp and the signal to be conducted with conductive gel in an intricate and time consuming process. Emotiv claim that their system is able to perform what appear to be better results than the state of the art in BCI with only dry electrodes and without any real care in the electrode placement.
Another problem with the portrayal of the emotive device is that it seems to operate very differently from traditional cursor control BCI systems. Cursor control generally requires motor imagery. This means imagining a movement in a part of your body which “lights up” a part of your brain and programming a computer to recognise when you are imagining this movement. This requires pinpointing signals from a part of the brain called the motor cortex…
Besides being relatively small, this problem is made worse by the fact that the nerves that send and receive motor information around the body are not evenly distributed. This if you have not seen it before is called a homunculus and is the widely regarded and long held model of how psychologists believe the motor system is distrbuted proportionally around the body (this is why contemporary BCIs revolve around imagined hand movement)…
And just for fun recent research has even suggested that a sensory homunculus may even look a bit more like this…
My sincere apologies for inflicting that last image on you. No comment on the implications of that with regard to BCI though, just no, final answer.
My point is that BCI systems as they are known today require us to imagine things that can be very hard to identify alone let alone program a computer to decode so the process as it stands is considerably more complicated than imagining a computer doing something and “poof” it happening. Unfortunately the signals given off when we think are far more complex than any TV or radio signal and can’t be decoded in the same way. I’m not saying it will forever be impossible just that development will take time, progress will be made in baby steps and we may well be far further behind the point in the developmental process than this video suggests we are. It would be a great shame for the importance of future research to be diminished by a popular perception that “someone else did all that years ago”. I really don’t want to throw sand on anyones bonfire I just believe you have the right to be well informed.
These three factors that I mentioned, in combination with a lack of any published work make myself and all of the experts I have spoken to, very reluctant to believe that this device will prove as effective as it is billed. The main concern of the critics to date is that the emotiv may actually use facial muscle movements rather than brain waves to provide the minimal control it allows. This would make the headset a rather obscure and ineffective way of picking up these signals (which would be EMG signals). One professor of Engineering working within the BCI field who I won’t name has described such devices as “Head Computer Interfaces rather than Brain Computer Interfaces”. We won’t know either way until someone does a controlled study and makes the results public. If the critics are wrong however this will come to be seen as a commercial breakthrough of epic proportions.
Regardless of whether the discoveries made by Emotiv are as grand as they sound or not the team at Emotiv certainly deserve credit for their work so far, their product is a step towards the hopes and dreams of all that have stepped foot in the field and we are all looking forward to seeing some published results by the company and hearing how this research develops.
Until then however, I won’t be splashing out the the $500 (just slashed from $2500) for the research edition of the product and nor will I be buying the somewhat cheaper consumer model just yet. I will however be keeping a close eye on the progress of this product as it does seem to be the most advanced product in a field just gagging for it’s bubble to burst.
For the record, I am only just beginning to dip my toe in to the deep dark lake that is this field so BIG respect to Emotiv for what they have achieved and all the other people working on mind bending projects like this. If I have learned one thing from seeing inside this field of research it is that it is going to be a very hard nut to crack.
If you are interested in this product this is their site and there is an interesting though now somewhat dead public google wave for users and developers of the emotiv product here.Follow Simon on Twitter, Facebook, Google+, RSS, or join the mailing list.
Cookie ComplianceThis site contains cookies. If you have ever used the internet before then you probably knew that already and ate them long before you arrived here. If you are allergic to cookies please leave now.