Home > On-Demand Archives > Microtalks >
Pre-screening Oral Tongue Lesions using TinyML
Mohammed Zubair Shamim - Watch Now - EOC 2022 - Duration: 06:21
Pre-screening Oral Tongue Lesions using TinyML
Mohammed Zubair Shamim
Diagnosing Oral Cavity Cancer (OCC) in its initial stages is an effective way to reduce patient mortality. However, current pre-screening solutions are manual and the resultant clinical treatment isn’t cost-effective for the average individual, primarily in developing nations. We present an automated and inexpensive pre-screening solution utilizing Artificial Intelligence (AI) deployed on embedded edge devices to detect benign and pre-malignant superficial oral tongue lesions. The proposed machine vision solution utilizes a clinically annotated photographic dataset of 9 types of superficial oral tongue lesions to retrain a MobileNetV2 neural network using transfer learning. In this approach, we also utilized TensorFlow Lite for Microcontrollers to quantize a 32-bit floating-point (float32) precision model into an 8-bit integer (int8) model for deployment on power and resource-constrained OpenMV Cam H7 Plus embedded edge device.
1 / 2
Please log in or create an account to test your knowledge and see the answers.
What deployment approach does Dr. Mohammed Zubair Shamim present for pre-screening oral tongue lesions on low-power embedded devices?
A
Use a TinyML pipeline: build a clinically annotated dataset of nine lesion types, train via Edge Impulse, quantize the model, and deploy it on an OpenMV Cam H7 Plus for real-time inference.
B
Train a large float32 MobileNetV2 on cloud servers and stream clinic images to the cloud for inference to avoid device constraints.
C
Rely on manual pre-screening guided by printed reference charts on the device to help non-specialists triage patients.
D
Implement a purely rule-based image processing algorithm on the OpenMV Cam H7 Plus (no ML) to avoid quantization and ML complexity.
E
Use a proprietary mobile app that uploads patient photos to a central server for batch processing without any edge inference.
