Architecture Overview

Technical Documentation

Detailed guide for clinical analysts and developers on how to interact with the AlzDetect system.

System Overview

AlzDetect is built with a decoupled architecture: a **FastAPI** backend handling the deep learning inference and a **Next.js** frontend for the diagnostic dashboard.

Inference Engine

Python-based environment running TensorFlow and Vit-Keras for high-performance MRI processing.

Diagnostic API

RESTful endpoints for file upload, disease classification, and attention map retrieval.

Diagnostic Protocol

01

Image Normalization

Inputs must be resized to 224x224 and normalized to a range of 0 to 1 before entering the transformer patches.

02

Attention Heatmapping

The attention maps are generated via a 16-layer attention rollout and overlayed with a customized JET colormap.

03

Decision Support

Results are presented with classification probabilities across four dementia stages.

Diagnostic Prototype Warning

AlzDetect is a technical prototype designed for demonstration purposes. It is not intended for clinical use or primary diagnosis without verification by a certified medical professional.