AI-Powered Medical Imaging

SonoX — Intelligent Ultrasound Image Analysis

SonoX leverages advanced deep learning to analyze ultrasound images, assisting clinicians in accurate lesion detection, boundary delineation, and malignancy classification across multiple body regions.

Quick Start Guide

Select a body region to analyze (e.g., Lymph Node, Breast, Thyroid)
Upload your ultrasound image or select from reference samples
Click "Start Analysis" to view AI segmentation and classification results

Research Institution

Developed by Prof. Michael Ying's research group at The Hong Kong Polytechnic University. For research purposes only.

Analysis Workspace

Select a body region, then upload or choose an ultrasound image to begin AI-powered analysis

Reference Samples

Click any sample to load it

Upload Image

Drag and drop or click to upload

Drop your image here

or click to browse files

AI Segmentation Result

Model prediction overlay

Upload an image and click Start Analysis

Red overlay = AI-predicted lesion

AI Lesion Classification

Benign vs. malignant probability prediction based on segmentation

Run analysis first

Classification results will appear after segmentation

About

An AI-powered ultrasound analysis platform for lesion segmentation and malignancy classification

What is SonoX?

An AI-powered platform integrating deep learning models for ultrasound lesion segmentation and malignancy classification
Assists clinicians in identifying and delineating lesion boundaries with high precision
Covers lymph nodes, breast, thyroid, and prostate ultrasound imaging

Segmentation

Based on UNet architecture with semi-supervised learning (Switch framework)
Trained with as little as 5% labeled data while maintaining competitive performance
Supports multi-region ultrasound segmentation across 6 clinical datasets

Classification

Vision Transformer (ViT) extracts deep visual features from lesion regions
Radiomics neural network captures handcrafted morphological and texture features
Dual-branch classification head fuses ViT and radiomics features for benign vs. malignant prediction

Publication

Published in IEEE Transactions on Neural Networks and Learning Systems (TNNLS), 2026
Validated on 6 clinical datasets from multiple medical centers
Demonstrates strong generalization across diverse patient populations

Important Medical Disclaimer

SonoX is designed for research and educational purposes only. It is not a certified medical diagnostic tool. All AI-generated results are for reference only and must not replace professional medical judgment, clinical examination, or established diagnostic procedures. Always consult qualified healthcare professionals for medical decisions.