A dynamic mechanism of Alzheimer based on artificial neural network

In this paper, we provide another angle to analyze the reasons why Alzheimer Disease exists. We analyze the dynamic mechanism of Alzheimer Disease based on the cognitive model that established from artificial neural network. We can provide some theoretic explanations to Alzheimer Disease through the analyzing of this model.

💡 Research Summary

The manuscript attempts to explain Alzheimer’s disease (AD) by invoking a computational model of human cognition derived from artificial neural network (ANN) theory. Building on Cheng’s 2010 “cognitive development model,” the author treats the brain as a fully‑connected network of N(t) neurons whose “computational complexity” C(t) and “cognitive depth” D(t) evolve over time. Two central equations are presented: (1) an double‑exponential decay function for C(t) and (2) an expression derived from the binomial theorem for the total number of network states, which the author equates with computational complexity. The time variable t is measured in months, and a constant h = 0.0001/15 is introduced without empirical justification.

In the “normal” scenario, the model predicts that computational complexity peaks at 300 months (≈25 years) and then declines, while cognitive depth peaks later at 600 months (≈50 years). The two curves intersect, after which cognitive depth follows the same declining trajectory as computational complexity. This qualitative picture is illustrated in Figure 1, but no real neurophysiological data are used to calibrate the curves.

The core of the paper consists of four hypothetical damage scenarios:

1. Sudden neuron loss – a 5 % reduction in N(t) at 600 months. The author shows that the complexity curve drops abruptly but then continues its original decline, resulting in a 1000‑month complexity equivalent to that at 118 months (Figure 2).

2. Accelerated functional decay – the network’s efficiency is assumed to decay exponentially faster, leading to a modified equation (6). The resulting curve (Figure 3) places the 1000‑month complexity at the level of a 97‑month brain.

3. Combined loss and decay – both the 5 % neuron loss and accelerated decay are applied simultaneously, producing an even steeper decline (Figure 4) where the 1000‑month complexity corresponds to an 85‑month level.

4. Sustained neuron attrition – a continuous loss of 0.05 % of neurons per month after the 300‑month peak. This produces a dramatically steep curve (Figure 5), with the 1000‑month complexity equivalent to that of a 45‑month-old.

From these simulations the author argues that the brain’s neural network is intrinsically stable, but chronic, low‑grade damage (e.g., from smoking, alcohol, certain drugs) can precipitate a rapid collapse of computational capacity, effectively reducing an adult’s cognitive performance to that of a child. This, the paper suggests, may underlie the emergence of AD. The author further posits that early control of such damaging factors could slow disease progression.

The discussion acknowledges the simplicity of the model (a fully‑connected network with uniform neuron behavior) and calls for deeper research to refine the approach and possibly develop preventive strategies. References are limited to Cheng’s own prior work and Piaget’s developmental psychology texts.

Critical appraisal reveals several major shortcomings:

• Over‑simplified architecture – Real cortical networks are sparse, hierarchical, and exhibit diverse neuron types and synaptic dynamics. Modeling the brain as a fully‑connected homogeneous network ignores these essential features.
• Arbitrary parameterization – Constants such as α, τ, h, and the rates of neuron loss are chosen without empirical grounding. No fitting to neuroimaging, electrophysiology, or neuropathological data is presented.
• Lack of validation – The paper provides only illustrative plots; there is no comparison with longitudinal cognitive test scores, brain volume trajectories, or biomarkers (e.g., amyloid‑β, tau) from actual AD cohorts.
• Misinterpretation of “computational complexity” – Equating the number of possible network states (derived from the binomial theorem) with cognitive ability is conceptually flawed. Cognitive performance depends on functional connectivity, information integration, and plasticity, not merely on combinatorial state counts.
• Neglect of disease‑specific mechanisms – AD pathology involves protein aggregation, neuroinflammation, vascular factors, and selective regional vulnerability, none of which are captured by a generic decline in neuron count or network efficiency.
• Insufficient methodological detail – The derivations of equations (1)–(6) are opaque, and the numerical simulations lack description of initial conditions, discretization steps, or software tools used.

In summary, while the manuscript offers an imaginative attempt to link ANN‑based mathematical modeling with Alzheimer’s disease, the current formulation is too abstract, under‑parameterized, and disconnected from empirical neuroscience to provide meaningful insight. Future work would need to integrate realistic network topologies, data‑driven parameter estimation, and validation against clinical and biological markers of AD to move beyond speculative speculation.