Discover

What changed

Task-trained deep networks predicted both V1 and IT better in these studies. They still leave nearly half of V1's explainable variance unexplained, only one supervised network was tested for IT so supervision is confounded with architecture and data, and the data come from small image sets and few animals.

What changed

The brain–symptom associations and the cluster structure both fail under appropriate validation. This shows the method cannot support the claim — not that biological subtypes of depression do not exist.

What changed

The relationship is disease-specific rather than a property of tau. The design is cross-sectional so it cannot show spread over time, the groups are small, and the PET ligand binds aggregated tau with known off-target signal.

What changed

Thickness is the dominant component, and it replicates across four samples and three countries. MRI cannot show the cellular processes behind it, and the accelerated design combines age cohorts rather than following individuals from 7 to 29.

What changed

A meaningful amount is present at diagnosis and continues faster in those with cognitive impairment. The relationships are correlational, so thinning is not shown to cause cognitive decline, and 18 months is short relative to the disease course.

What changed

Two modalities agree within one sample. The participants are high-functioning children and adolescents, so this does not extend to adults or to lower-functioning individuals, and a cross-sectional design cannot say whether these differences precede or follow the behavioural phenotype.

What changed

This study places mPFC activity on a path between anhedonia and function. Sample size is small and the design is cross-sectional, so it is a mechanistic hypothesis with imaging support, not a confirmed causal chain.