Where the Lesion Sits Adds a Little – and Only at the Dementia End
- In 1253 rural Chinese adults aged 60 and over (median age 69, IQR 66–72; 59.06% women; median 3 years of schooling, IQR 0–6), six of eleven fibre tracts held a cross-sectional association with dementia after false discovery rate correction: forceps minor OR 1.36 per 1 ml (95% CI 1.13–1.64, corrected p = 0.007), corticospinal tract 1.13 (1.05–1.21, p = 0.007), forceps major 1.12 (1.02–1.24, p = 0.040), anterior thalamic radiation 1.09 (1.03–1.15, p = 0.017), inferior fronto-occipital fasciculus 1.09 (1.02–1.17, p = 0.030), superior longitudinal fasciculus 1.03 (1.01–1.06, p = 0.033).
- Total lesion volume was also associated with dementia (OR 1.008, 95% CI 1.00–1.01, corrected p = 0.033), but location did more than add to it: on DeLong's test the models carrying anterior thalamic radiation, corticospinal tract or forceps minor each discriminated significantly better than the model carrying total lesion volume. That comparison is what the authors mean by a strategic location.
- The entire signal sat at the dementia end. Neither total nor tract-specific lesion volume was associated with subjective cognitive decline or with mild cognitive impairment once the twelve comparisons were corrected; the smallest corrected p in those columns was 0.065.
- Discrimination gained little in size, though not nothing in statistical terms. Age, sex and education alone separated dementia from normal cognition at AUC 0.792 (95% CI 0.716–0.869); adding total lesion volume gave 0.801 (0.725–0.877); the best single tract, forceps minor, gave 0.822 (0.752–0.892). On DeLong's test four tracts discriminated significantly better than demographics alone, and three better than demographics plus total volume. Across all thirteen models, mild cognitive impairment versus normal cognition stayed between 0.683 and 0.687.
- The design was cross-sectional and only 28 participants (2.23%) had dementia, which is the stratum carrying every one of the six tract associations.
Nearly every brain MRI in an older patient comes back with a line about white matter hyperintensities, and almost none of those lines changes what happens next in the consulting room. A group at Shandong Provincial Hospital and Shandong First Medical University asked the question differently: instead of one lesion volume, measure where the lesions sit, tract by tract, in a population sample rather than a memory clinic. They used 1253 participants from the MIND-China study in Yanggu County, western Shandong, scanned between 2018 and 2020. The result is worth reading precisely because half of it is negative.
The tract map
Lesions were segmented from 3D FLAIR images with the Lesion Prediction Algorithm, normalised to MNI152 space and overlaid on the Johns Hopkins probabilistic fibre-tract atlas, yielding eleven regions of interest. Total lesion volume rose across the cognitive spectrum: median 12.01 ml in normal cognition (n = 669), 15.56 ml in subjective cognitive decline (n = 270), 17.10 ml in mild cognitive impairment (n = 286) and 27.20 ml in dementia (n = 28).
Against normal cognition, adjusted for age, sex and education, six tracts survived correction across twelve tests. Forceps minor was the strongest per millilitre at OR 1.36 (95% CI 1.13–1.64), followed by corticospinal tract 1.13 (1.05–1.21), forceps major 1.12 (1.02–1.24), anterior thalamic radiation 1.09 (1.03–1.15), inferior fronto-occipital fasciculus 1.09 (1.02–1.17) and superior longitudinal fasciculus 1.03 (1.01–1.06). The temporal part of the superior longitudinal fasciculus carried the largest raw odds ratio of the whole table, 2.14 (1.06–4.34), and it was the one association that did not survive correction (corrected p = 0.052) – a clean illustration of what a wide interval on a rarely affected tract is worth. Anatomically the two leading tracts are the ones that would be predicted: anterior thalamic radiation carries the prefrontal-subcortical and Papez connections, forceps minor carries frontal interhemispheric traffic.
Where the map goes quiet
Below the dementia threshold the picture collapses. Neither total nor tract-specific lesion volume was associated with subjective cognitive decline or with mild cognitive impairment after correction; several raw p values were under 0.05 but the smallest corrected value in those two columns was 0.065. Among 955 participants with normal cognition or mild cognitive impairment, anterior thalamic radiation (OR 1.03, 95% CI 1.00–1.07, p = 0.083) and corticospinal tract (1.04, 1.00–1.09, p = 0.076) were marginal at best for the amnestic subtype, and nothing approached significance for the non-amnestic one.
The discrimination table says the same thing more bluntly. Demographics alone – age, sex, education – reached AUC 0.792 for dementia versus normal cognition. Total lesion volume added 0.009; the best tract added 0.030. Neither increment is statistically empty, and this is the paper's own headline: on DeLong's test four tracts beat the demographic model, and anterior thalamic radiation, corticospinal tract and forceps minor beat the model that already carried total lesion volume. The quarrel is with the size of the gain, not with its significance. For mild cognitive impairment versus normal cognition all thirteen models sat between 0.683 and 0.687, so the imaging moved nothing at the stage where an early marker would actually be useful.
One mismatch of presentation needs naming. The abstract states that in dementia-free participants higher lesion volumes in anterior thalamic radiation and forceps major were associated with lower memory z-scores, and volumes in corticospinal tract, forceps major, inferior fronto-occipital fasciculus and superior longitudinal fasciculus with lower verbal fluency – flagged there, in fairness, as uncorrected p < 0.05. The footnote to Table 4 states that no association in that table survived false discovery rate correction, and the results text puts the best of them, forceps major with verbal fluency, at corrected p = 0.084. The Conclusions then describe these associations as independent across the cognitive spectrum, and that is the step the table does not support. The table is the stricter and more defensible reading, and this analysis follows the table.
What changes in the consulting room
Read a lesion report in a patient with subjective complaints or mild cognitive impairment as a vascular-risk signal, not as a stage marker – in this sample it did not separate those states from normal cognition once multiple comparisons were corrected for. When the burden is heavy anteriorly, in forceps minor and anterior thalamic radiation, the prefrontal-subcortical route the authors describe is the plausible anatomy behind the memory and verbal-fluency profile; that is a reason to attend to vascular risk on general clinical grounds, not a finding of this study, which measured nothing about treatment and cannot order events in time.
The sample itself is part of the message. This is a rural, socioeconomically disadvantaged, under-studied population: median 3 years of schooling, 35.35% with no formal education, hypertension in 67.45% of the whole sample with no significant difference across cognitive groups (p = 0.081), and APOE ε4 in 15.25% likewise flat across groups (p = 0.705). In a cohort that uniformly vascular-loaded, lesion topography separated cognitive states only at the far end. Funding came from Chinese national and provincial programmes with additional Swedish grants to one of the senior authors at Karolinska Institutet; the authors declare no competing interests.
The lesion map earns its keep at the dementia end of the spectrum and goes quiet exactly where an early marker would be worth having.
The design is cross-sectional, so neither temporal nor causal ordering follows from it. All six tract associations rest on 28 participants with dementia, and the authors note that lesions were rare in some functionally relevant tracts, limiting power. The abstract reports domain-level associations at uncorrected p < 0.05 while the footnote to Table 4 states that none survived correction; this analysis follows the table. The sample comes from one rural county in northern China.