AI-assisted Thyroid Cancer Diagnosis System
About it
The Thyroid FNA AI-Assisted Diagnostic System is designed for fine-needle aspiration (FNA) cytology of thyroid nodules.
It focuses on benign–malignant differentiation and subtype classification, accurately recognizing papillary carcinoma and follicular lesions.
Developed to address the challenges of subjective diagnosis and difficult follicular lesion distinction, it shortens FNA analysis time and provides objective pathology evidence for low-risk nodule follow-up.
FNA-Specific Optimization
Optimized for thyroid FNA smears (manual or liquid-based), accurately handling sparse and overlapping cells.
Precise Lesion Differentiation
Identifies PTC, SFN, BFN, and AUS with high accuracy, resolving challenges in follicular lesion distinction.
High Sensitivity & Specificity
95% sensitivity and 95% accuracy for SFN vs. BFN; 100% full-field coverage without missed areas.
Top-Tier Clinical Data
Trained on 10,000+ biopsy-verified slides from 10+ top-tier hospitals with a 2:1 negative–positive ratio.
Highlights nuclear grooves and ground-glass nuclei for explainable AI-based diagnosis.
Compatible with slide formats and image outputs from various manufacturers and scanner systems.
Rapid Structured Report
Generates a full structured report in under 1 minute — ready for printing or LIS integration.
Recognizable Lesion Types
| Category | Detected Types | Description |
| Malignant Lesions | Papillary Thyroid Carcinoma (PTC), Suspicious for Malignancy (SFM) | Accurately identifies malignant cytological features and papillary carcinoma subtypes. |
| Borderline Lesions | Suspicious Follicular Neoplasm (SFN), Atypia of Undetermined Significance (AUS) | Differentiates follicular-patterned lesions with high diagnostic precision. |
| Benign Lesions | Benign Follicular Nodule (BFN) | Recognizes benign follicular nodules and provides pathology basis for follow-up recommendations. |
Powering global digital pathology with performance, scale, and trust
The Numbers Behind Our Innovation
Network and GPU Server Requirements
| Parameter Category | Minimum Requirement | Recommended Configuration | Remarks |
| Local Network Bandwidth | Between server and client ≥100Mbps | Between server and client ≥1Gbps (preferably optical) | Core switch supports ≥10Gbps uplink |
| Network Latency | ≤50ms (between LAN nodes) | ≤50ms (between LAN nodes) | Supports link aggregation (LACP) for improved stability |
| Network Topology | Supports star/tree topology | Supports redundancy (dual-core switch) | Key nodes (e.g., database server) require dual-network binding |
| Interface Type | Client RJ45 (10/100BASE-T) | Server RJ45 (1000BASE-T) + Optical port (optional) | Optical interface supports LC/SC connectors, transmission distance ≤5km |
| Parameter Category | Minimum Requirement | High-Performance Configuration (e.g., AI Training / Rendering) | Remarks |
| GPU Model | NVIDIA 2080Ti ×1 | NVIDIA A4000 ×1 | |
| GPU Memory | 11GB GDDR6 (ECC optional) | 16GB HBM2e (ECC required) | |
| GPU Quantity | Single GPU per host | Single GPU per host | |
| CPU | Intel Xeon E-2274G (4 cores, 8 threads, 3.8GHz) | Intel Xeon E-2274G (8 cores, 16 threads, 3.8GHz) | Main frequency ≥3.0GHz, cache ≥24MB |
| Memory | 32GB DDR4-2666 (ECC) | 64GB DDR5-4800 (ECC Registered) | |
| Storage Interface | 1×NVMe SSD (PCIe 3.0) | 2×NVMe SSD (PCIe 4.0, RAID 0) | Supports hot-swap, continuous read/write ≥3000MB/s |
| Power & Cooling | 900W 80+ Gold PSU, air cooling | 1600W 80+ Platinum PSU, liquid cooling system | |
| OS Compatibility | Supports Ubuntu 22.04 server version | Supports Ubuntu 22.04 server version | Docker pre-installed (ready for container deployment) |
