Employment Project
Jan 2019 - Jan 2020
CADtnet
Edge TFLite Breast Cancer Ultrasound Triage App
Organization: xtLytics LLC
Key Result & Impact Benchmark
Enabled real-time, offline breast lesion screening with high sensitivity across low-resource clinics in LMICs.
Technical Overview
CADtnet is an edge-optimized mobile AI application built to triage breast cancer lesions from ultrasound scans in Low-Middle Income Countries (LMICs) where specialized radiologists are scarce. Leveraging custom lightweight convolutional neural networks (VGG-19, MobileNet, ResNet, U-Net++), the model was quantized into TensorFlow Lite (TFLite) to run real-time inference directly on Android smartphones without requiring active internet connectivity or cloud servers. The system was field-tested and deployed in Guadalajara, Mexico.
Key Technical Highlights
- Trained on a curated clinical dataset of ~6,500 ultrasound scans from scratch to deployment.
- Achieved sub-second edge inference on mid-range Android mobile devices via 8-bit TFLite quantization.
- Field-deployed in Guadalajara, Mexico to support clinical health workers in resource-constrained environments.
Technologies & Frameworks
TFLite
VGG-19
MobileNet
ResNet
U-Net++
Android
Python