Advancing ADHD diagnosis: How Qbtech built a mobile AI assessment Model Using Amazon SageMaker AI

This article describes how Qbtech built QbMobile, a smartphone-based ADHD assessment tool using Amazon SageMaker AI and AWS Glue, reducing feature engineering time from 2 days to 30 minutes.

• QbMobile uses smartphone cameras and motion sensors for clinical-grade ADHD diagnosis
• Binary LightGBM model processes 24 features from face tracking, head movement, and error patterns
• Amazon SageMaker AI parallel processing reduced feature engineering by 96% through multiprocessing
• AWS Glue standardized heterogeneous smartphone data across different device types
• Model achieved 85.7% sensitivity, 74.9% specificity, and 73.2% PR-AUC
• Results delivered in under one minute from data collection to inference
• Enables 100% remote diagnostic process for patients with logistical barriers
• Deployment uses Docker, ECR, GitHub Actions, and Terraform on AWS infrastructure
• Comprehensive security includes encryption, multi-factor authentication, and access logging

Qbtech's cloud-based approach democratizes ADHD assessment access globally while maintaining clinical standards and enabling rapid model iteration.