Quantum-assisted Alzheimer's risk from MRI genetics and CSF biomarkers

ALZQAPI · MRI Genetics Research Platform

Download paper

Kryptur Research presents a quantum kernel support-vector pipeline that encodes 68-region cortical MRI thickness, APOE ε4 dosage, CSF amyloid/tau biomarkers, and MMSE into an 8-qubit ZZFeatureMap Hilbert space — estimating preclinical Alzheimer's risk from state fidelity on IBM ibm_fez.

10.5281/zenodo.20944392Raja RamTauNet 1.0
QPU validated AUC
0.9167
Noise-normalised quantum kernel on IBM ibm_fez
Encoding qubits
8
ZZFeatureMap · reps=2 · full entanglement
Cortical regions
68
MRI thickness features PCA-reduced before encoding
Training cohort
200
Synthetic patients · precomputed kernel SVM

Recent news

ALZQAPI: Quantum Kernel SVM for Preclinical Alzheimer's Risk from MRI Genetics and CSF Biomarkers

Research Paper
Raja Ram M · Main Author · Kryptur Quantum R&D · Zius Research Center

This report documents the ALZQAPI pipeline: 73-dimensional patient vectors (68 cortical thickness values, APOE ε4 count, CSF Aβ42, t-tau, p-tau181, MMSE) are PCA-reduced to eight components, encoded via Qiskit ZZFeatureMap, and classified with a precomputed quantum-kernel SVM. The deployed simulator kernel achieves AUC 0.67; noise-normalised ibm_fez QPU validation reaches AUC 0.9167.

MRI GeneticsQuantum KernelAlzheimer's

QPU-validated kernel performance on IBM ibm_fez

Figure 1
Simulator vs raw QPU vs noise-normalised QPU AUC comparison
ibm_fezAUC

Kernel variant cross-validation across training pipelines

Benchmark
3-fold CV AUC for simulator, scaled, and QPU kernel paths
SVMZZFeatureMap
Open access archive on ZenodoRead more research on our hub

Stay up to date

with the latest from

Kryptur Research on

LinkedIn

Results & figures

Table 1. Kernel paths and 3-fold cross-validation AUC
Kernel pathBackend3-fold CV AUCDeployment status
Simulator kernelStatevector (noiseless)0.67Production serving
Raw QPU kernelIBM ibm_fez0.75Validation only
Noise-normalised QPU kernelIBM ibm_fez + ZNE scaling0.9167Research benchmark
Table 2. Patient biomarker inputs and encoding pipeline
Feature groupDimensionsDescription
MRI cortical thickness68Regional thickness values (mm) from structural MRI parcellation
APOE ε4 count1Genetic risk allele dosage (0–2)
CSF biomarkers3Aβ42, total tau, phosphorylated tau (p-tau181) in pg/mL
MMSE1Mini-Mental State Examination cognitive score (0–30)
PCA + scale → ZZFeatureMap8 qubits73-dim vector reduced to 8 components, encoded with reps=2 full entanglement
AUC comparison of simulator, raw QPU, and noise-normalised quantum kernel paths
Figure 1. 3-fold cross-validation AUC for the deployed simulator kernel (0.67), raw ibm_fez QPU kernel (0.75), and noise-normalised QPU kernel (0.9167) — demonstrating quantum advantage after noise scaling on real hardware.
Cross-validation AUC across quantum kernel training variants
Figure 2. Kernel variant comparison across training pipelines — simulator, scaled, and Qiskit v2 implementations evaluated on the 200-patient synthetic cohort.
Scaled quantum kernel AUC benchmark results
Figure 3. Scaled quantum kernel benchmark — feature scaling and PCA loadings applied before ZZFeatureMap assignment improve class separation in Hilbert space.
Second-generation kernel comparison with ibm_fez validation
Figure 4. Second-generation kernel comparison — v2 Qiskit runtime path with updated support-vector selection and ibm_fez hardware validation cohort.

Open access · ALZQAPI Research Initiative

Kryptur OU · doi:10.5281/zenodo.20944392 · Main Author Raja Ram M · Zius Research Center · MRI Genetics

Download full report

Tools + Code

TauNet 1.0 API

Live Alzheimer's risk console — single-patient scoring, batch CSV upload, quantum kernel heatmap, and SHAP feature chart.

View project →

Research Paper

ALZQAPI engineering report — quantum kernel SVM, ZZFeatureMap encoding, ibm_fez QPU validation, and browser risk-assessment frontend.

View project →

api.py

Flask REST API — POST /api/v1/alzheimers-risk with SHAP feature attribution, batch CSV ingestion, and precomputed quantum kernel SVM.

View project →

quantum_kernel_qiskit.py

Qiskit ZZFeatureMap statevector kernel — |⟨φ(xᵢ)|φ(xⱼ)⟩|² fidelity between 8-qubit encoded patient biomarker vectors.

View project →

preprocess.py

73-dimension feature pipeline — 68 cortical thickness regions, APOE ε4 dosage, CSF biomarkers, MMSE, PCA reduction, and scaling.

View project →

VQPP Surgical Planning

Complementary variational quantum path planning for robotic surgery — GPU-accelerated 3D anatomy to QUBO on IBM Kingston.

View project →