Academic Jobs - Home of Higher Ed Logo

Research spotlight: Isobel French

10 min read0 Ansichten

At a glance

A study co-authored by Isobel French at National Central University found that long-range temporal correlations in resting EEG tracked faster motor and cognitive decline in Parkinson's disease over two years.

A close up of a book with writing on it
Photo by Brett Jordan on Unsplash

A ten-minute brainwave recording, taken while a patient sits still with their eyes closed, appears to carry information about how quickly Parkinson's disease will advance over the following two years. That is the substance of a study published in Movement Disorders and co-authored by Isobel French, a doctoral student and research assistant at National Central University in Taiwan.

The team tracked 63 people with Parkinson's disease and 58 healthy controls of matched age and sex. Everyone had an electroencephalogram (EEG), the scalp recording that picks up the summed electrical chatter of billions of neurons, plus a standard clinical work-up. Two years later, the patients were scored again. The way their brain rhythms had waxed and waned at baseline — a property called long-range temporal correlations, or LRTC — lined up with how much their motor and cognitive scores had slipped.

The timing matters. Parkinson's disease has no validated blood test or scan that tells a neurologist whether a newly diagnosed 60-year-old will be walking unaided in five years or needing help with buttons in two. Doctors adjust drug regimens by trial and error, and clinical trials of disease-modifying therapies struggle to show a benefit because everyone in the placebo arm declines at a different rate. A modelling study published in The BMJ in 2025 projected that global cases will reach 25.2 million by 2050, up from roughly 11.8 million in 2021. A cheap, widely available readout of progression speed would matter at that scale.

What the study actually did

Participants were recruited through a collaboration involving National Central University's cognitive neuroscience researchers and clinical neurologists in the Chang Gung hospital system. Each person underwent resting-state EEG and a battery of assessments: the motor examination section of the Unified Parkinson's Disease Rating Scale (UPDRS part III), the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA). The two cognitive scales are the standard pencil-and-paper screens for orientation, memory, attention and language.

Of 116 patients initially enrolled, 63 returned for the two-year follow-up. That attrition is one of the study's genuine weaknesses, and worth holding onto while reading the rest.

The analysis itself did not look at raw brainwaves. It looked at the envelope of specific frequency bands — theta (roughly 4 to 8 hertz, the slow, drowsy rhythm), alpha (8 to 12 hertz, the idling rhythm that dominates when eyes are shut) and gamma (above 30 hertz, the fast, low-amplitude activity that is hardest to capture through the skull). In each band and at each electrode site, the researchers measured how strongly the amplitude at one moment depended on the amplitude seconds earlier.

Reading a memory in a brain rhythm

Flip a fair coin a hundred times. Knowing the last five results tells you nothing about the next one. That independence is what statisticians call white noise, and it corresponds to a number of 0.5. Long-range temporal correlations are the opposite: the signal's own history bends what comes next, and it keeps bending it over tens of seconds. A tide has that kind of memory. A coin toss does not.

The tool the team used to measure it is detrended fluctuation analysis, or DFA. The method, first applied to brain oscillations by Klaus Linkenkaer-Hansen and colleagues in 2001, works like this: take the amplitude envelope, add the values up cumulatively, then slice the resulting curve into windows of increasing size. Inside each window, fit a straight line and measure how far the actual data strays from it. Small windows produce small deviations; large windows produce large ones. Plot the deviation against window size on log-log axes and you get a straight line. Its slope is the DFA exponent. A slope near 0.5 means no memory. A slope between 0.5 and 1 means the rhythm remembers itself.

Critically, the exponent is not just a restatement of how loud a rhythm is. Work from twin studies and large resting-state cohorts found that DFA exponents are largely independent of oscillation power, stable across sessions, and roughly 60 percent heritable. It captures something about how a network organises itself over time, not merely how hard it is firing.

A close up of a book with a page in it

Photo by Brett Jordan on Unsplash

Three rhythms, three different jobs

Compared with healthy controls, patients with Parkinson's disease showed higher LRTC in theta oscillations over the left parietal region (P = 0.0175) and lower LRTC in gamma oscillations over the centro-parietal region (P = 0.0258). The parietal lobe sits toward the back of the head and handles spatial attention and the integration of sensory information.

The more telling results came from the two-year change scores. Weaker gamma-band LRTC at baseline tracked a larger rise in UPDRS part III scores, the motor exam that rates tremor, rigidity, slowness and gait (Spearman ρ = −0.34, P = 0.0082). Stronger theta-band LRTC in the left parietal region was associated with what the team defined as rapid motor progression: an annual increase of at least 3 points on that motor exam (P = 0.0107). And parieto-occipital alpha LRTC moved in step with cognitive change, correlating positively with shifts on both the MMSE and the MoCA across the same two years (ρ = 0.27 to 0.38, P = 0.0037 to 0.0452).

Read plain: a slow rhythm that holds onto its own past too tightly, and a fast rhythm that has lost its grip on its past, both marked patients whose bodies would deteriorate faster. The alpha finding ran the other way, tying a more coherent resting rhythm to a better-preserved cognitive trajectory.

The authors are careful with their own conclusion. They write that LRTC patterns "potentially predict rapid progression of both motor and non-motor manifestations in PD patients, enhancing clinical assessment and understanding of the disease." Potentially is doing real work in that sentence.

Why a cheap test matters now

Parkinson's biomarker research has moved fast, but mostly in directions that are hard to scale. The alpha-synuclein seed amplification assay, which detects misfolded protein in cerebrospinal fluid, has high sensitivity and specificity for established disease. A 2025 review in The Lancet Neurology noted that these assays remain largely confined to research settings and require a lumbar puncture, which is not something you repeat every six months to check on progression. A positive result tells you the pathology is present. It does not tell you how fast it is moving.

EEG is the opposite kind of instrument. The hardware costs a fraction of an MRI scanner, the recording takes minutes, it needs no injected tracer and it can be repeated without risk. Hospitals in low- and middle-income countries often have an EEG machine already. If a DFA exponent could be extracted from a routine resting recording and dropped into a neurology report, the practical reach would be far wider than anything requiring a specialised assay.

The East Asian picture sharpens the argument. The BMJ projections put China at 10.5 million cases by 2050 and the wider East Asia region at 10.9 million, the largest regional total anywhere, with population ageing accounting for 89 percent of the global increase. A tool that helps triage which patients need aggressive management soonest has an obvious home in health systems bracing for that volume.

What could go wrong

The first problem is arithmetic. Nearly half the original patient group never made it to the two-year assessment. People who drop out of Parkinson's studies are not a random sample; they are often the sickest, the most cognitively impaired, or those who have moved into residential care. That kind of loss can either inflate or deflate an association depending on who leaves, and no statistical correction fully repairs it.

Second, the correlations are real but modest. A Spearman coefficient of −0.34 means the gamma measurement explains a slice of the variance in motor decline, not the whole pie. No clinician should read a single DFA exponent as a prognosis.

Third, DFA exponents vary enormously between healthy people to begin with. In one cohort of more than 1,400 subjects, alpha-band exponents ranged from about 0.55 to 1.05. Any Parkinson's signal has to be detected on top of that natural spread, and the field has not settled on a universal cut-off that separates normal from abnormal.

Fourth, technical fragility. Scalp EEG picks up gamma activity poorly, because the skull blurs fast, low-amplitude signals and because scalp muscles generate their own high-frequency noise. Recordings taken with eyes open versus eyes closed also shift the exponents, and arousal level changes them further. A marker that moves when a patient is simply more alert than last time is a hard sell in a clinic.

Fifth, the design is observational over a single two-year window in one cohort. An association between a baseline measurement and later decline is not the same as a validated prediction rule, and the paper does not claim otherwise. Replication in an independent, multi-site sample is the test that has not yet been run. Until it is, a reasonable neurologist will treat this as a promising lead rather than a usable instrument. The frequency-band-by-region analysis also multiplies the number of comparisons being made, which raises the chance that some of the smaller effects will not survive replication.

a red car with black text

Photo by Nick Fewings on Unsplash

Who funded the work, and who did it

The published record lists support from the Chang Gung Medical Foundation. The article carries the copyright of the International Parkinson and Movement Disorder Society, which publishes Movement Disorders. The study was a joint effort between engineers of brain signals and the neurologists who see these patients in clinic, and the author list reflects that: Chih-Hong Lee, Chi-Hung Juan, Hsiang-Han Chen, Jia-Pei Hong, Ting-Wei Liao, Isobel French, Yen-Shi Lo, Yi-Ru Wang, Mei-Ling Cheng, Hsiu-Chuan Wu, Chiung-Mei Chen and Kuo-Hsuan Chang.

French's contribution sits inside a longer programme. She has co-authored earlier work from the same collaboration applying Holo-Hilbert spectral analysis, a nonlinear method for decomposing brain signals, to Parkinson's disease staging and to attentional dysfunction. Her 2025 doctoral dissertation at National Central University examined resting-state and attentional neural correlates of Parkinson's disease using EEG and nonlinear analytical methods, including machine-learning classification of patients against controls. This LRTC paper is a different angle on the same question: which features of a raw brain signal carry clinical information that a rating scale alone misses.

For early-career researchers, the paper is also a model of a particular kind of collaboration. A doctoral student with signal-processing expertise embedded in a clinical neurology group, working with longitudinal follow-up data that takes two years to accumulate, produced a result that neither the engineering lab nor the movement disorders clinic could have generated alone.

What would have to happen next

The near-term agenda is unglamorous and specific: repeat the analysis in an independent cohort with better retention, pre-register the frequency bands and electrode sites, and report whether the exponents hold up when the same patients are recorded twice on different days. If the signal survives that, the next test is whether adding an LRTC reading to a standard clinical assessment actually improves a doctor's judgment — a question that requires a prospective trial, not another retrospective correlation.

Further out, the more interesting possibility is trial design. Disease-modifying trials in Parkinson's fail partly because progression is measured by rating scales that shift with the examiner's mood and the patient's medication timing. A physiological marker that reads the network's own dynamics, recorded in ten minutes on hardware most hospitals already own, could enrich a trial with the patients most likely to decline fast enough for a drug effect to be visible. That would be a quieter contribution than a headline cure, and a more durable one.

  • Parkinson's disease
  • long-range temporal correlations
  • detrended fluctuation analysis
  • biomarkers
  • Isobel French
Porträt von Jarrod Kanizay
Über den Autor

Jarrod KanizayAutor ansehen

Academic Jobs In House Author

Visionary leader transforming academic recruitment with 20+ years in higher education.

Die Anerkennung :

Diskussionen

Sort von:

Seien Sie der Erste, der diesen Artikel kommentiert!

Du bist

Sie werden gebeten, sich anzumelden, bevor Ihr Kommentar veröffentlicht wird.

Browse nach Fakultät

Browse nach Thema

Frequently Asked Questions

🧠What are long-range temporal correlations in EEG?

They describe how much a brain rhythm's amplitude at one moment depends on its amplitude seconds earlier. A signal with no memory, like coin flips, scores about 0.5. A signal where large fluctuations tend to follow large fluctuations scores between 0.5 and 1. Long-range temporal correlations (LRTC) in neuronal oscillations were first characterised systematically by Linkenkaer-Hansen and colleagues and are thought to reflect networks sitting near a critical, highly adaptable state.

📈What did the Parkinson's study actually measure?

Sixty-three patients with Parkinson's disease and 58 age- and sex-matched healthy controls had resting-state EEG plus clinical assessments. The team applied detrended fluctuation analysis to the amplitude envelopes of theta, alpha and gamma oscillations across electrode sites. Patients were then rescored two years later on the UPDRS part III motor exam and on the MMSE and MoCA cognitive screens.

⚡Which brain rhythms predicted faster motor decline?

Two did, in opposite directions. Lower LRTC in centro-parietal gamma oscillations correlated with a larger two-year rise in UPDRS part III scores (Spearman rho = -0.34, P = 0.0082). Higher LRTC in left parietal theta oscillations was associated with rapid motor progression, defined as an annual UPDRS part III increase of 3 points or more (P = 0.0107).

🔍Did the study say anything about cognition?

Yes. LRTC in parieto-occipital alpha oscillations correlated positively with two-year changes on both the Mini-Mental State Examination and the Montreal Cognitive Assessment, with coefficients between 0.27 and 0.38 and P values from 0.0037 to 0.0452. In practical terms, patients whose alpha rhythm showed stronger temporal correlations at baseline tended to have a more favourable cognitive trajectory.

📐How is detrended fluctuation analysis calculated?

Take the amplitude envelope of a filtered oscillation, compute its cumulative sum, then divide that curve into non-overlapping windows of increasing length. Remove the local trend inside each window and measure the root-mean-square residual. The residuals grow with window size according to a power law, and the slope of the log-log fit is the scaling exponent. A slope near 0.5 indicates uncorrelated noise; a slope between 0.5 and 1 indicates persistent temporal correlations.

🎯How reliable is the DFA exponent as a measurement?

More reliable than you might expect for a scalp recording. Test-retest studies across sessions days apart found significant reliability, strongest with eyes closed. Twin work suggests roughly 60 percent of the variance between individuals is attributable to genetic factors. But the range across healthy people is wide, from about 0.55 to 1.05 in a cohort of more than 1,400, which is why a single reading cannot be used as a prognosis.

⚠️What are the main limitations of this study?

Of 116 enrolled patients, 63 completed the two-year follow-up, and dropouts in Parkinson's cohorts are often the sickest. The correlations are statistically significant but moderate. Scalp EEG captures gamma activity poorly because the skull blurs fast signals and scalp muscle adds high-frequency noise. The analysis is observational in a single cohort over one two-year window, so it identifies associations rather than establishing a validated prediction rule. The authors describe LRTC as something that potentially predicts progression.

🏛️Who funded the research and where was it done?

The published record lists support from the Chang Gung Medical Foundation. The work was a collaboration between National Central University's cognitive neuroscience researchers and clinical neurologists in the Chang Gung hospital system in Taiwan, and appeared in Movement Disorders, the journal of the International Parkinson and Movement Disorder Society.

🎓Who is Isobel French?

Isobel French is a doctoral student and research assistant at National Central University in Taiwan. Her 2025 dissertation examined resting-state and attentional neural correlates of Parkinson's disease using EEG and nonlinear analytical methods. She has co-authored earlier papers from the same collaboration, including work applying Holo-Hilbert spectral analysis to Parkinson's staging and a study of attentional dysfunction arising from network connectivity in the disease.

🧪How does EEG compare with alpha-synuclein seed amplification assays?

They answer different questions. A seed amplification assay detects misfolded alpha-synuclein in cerebrospinal fluid with high sensitivity and specificity, confirming that the pathology is present, but it requires a lumbar puncture and remains largely confined to research use. EEG is cheap, non-invasive and repeatable, and the LRTC approach attempts to read the rate of change rather than the presence of disease. The two could plausibly complement each other.

🌍Why does Parkinson's disease progression need a biomarker at all?

Because clinicians currently adjust treatment by observation and trial and error, and because trials of disease-modifying therapies struggle to detect benefit when participants decline at wildly different rates. Global cases are projected to reach 25.2 million by 2050, with population ageing driving 89 percent of the increase. A marker that sorts fast progressors from slow ones would help both individual treatment planning and trial design.

🛠️What would it take before this reaches a neurology clinic?

Independent replication in a multi-site cohort with stronger retention, pre-registered frequency bands and electrode sites, and evidence that the exponents are stable when the same patient is recorded on different days. After that, a prospective study would need to show that adding an LRTC reading to a standard clinical assessment actually improves a clinician's judgment. None of those steps has happened yet.