KFBIO Cervical Cancer AI Screening System

KFBIO’s Cervical Cancer AI Screening System applies deep convolutional neural networks to analyze liquid-based cytology specimens, automatically detecting and grading abnormal cells while filtering 50-70% of negative cases from manual review queues. The system received NMPA Class II certification in 2019, becoming among the first AI-powered pathology diagnostic tools commercially deployed in China.

Product Overview

Cervical cancer screening programs face a fundamental resource constraint: the shortage of qualified cytotechnologists and pathologists relative to screening population volumes. Manual microscopy of Pap smear or liquid-based cytology (LBC) samples remains time-consuming and subject to fatigue-related variability.

KFBIO developed the Cervical Cancer AI system through collaboration with Intel, building on the Intel-KFBIO Pathology AI Joint Laboratory established in 2018. The system integrates with KFBIO’s digital pathology scanners to create an end-to-end workflow from specimen digitization through AI-assisted diagnosis.

Clinical deployments have processed hundreds of thousands of screening cases across China, with projects expanding to Brazil for population-scale cervical cancer prevention initiatives.

Key Features

At 50–70% negative case filtering, KFBIO’s cervical cancer AI system lets two pathologists screen 100,000 annual cases — a workload that previously required four — using a NMPA Class II-certified deep learning pipeline for liquid-based cytology.

• Deep Convolutional Neural Network: Purpose-built DCNN architecture trained on retrospectively annotated cervical cytology samples from Chinese healthcare institutions

• Whole-Slide Analysis: Algorithms scan entire digitized slides rather than sampling limited fields, reducing false-negative rates from missed abnormalities

• Automated Cell Detection: Identifies suspicious cells throughout the specimen with coordinates marked for pathologist review

• Graded Lesion Classification: Automatically categorizes detected abnormalities by severity level, prioritizing cases requiring immediate attention

• Cell Quantity Assessment: Counts detected abnormal cells to support quantitative reporting standards

• Slide Quality Evaluation: Automated assessment of specimen adequacy before diagnostic analysis, flagging unsatisfactory preparations

• Error Correction Feedback: Built-in mechanism for pathologists to provide corrective input, enabling periodic algorithm updates

Clinical Performance

Validation studies and commercial deployments demonstrate meaningful workflow impact:

• Negative Case Filtering: AI eliminates 50-70% of negative cases from manual review queues
• Staffing Efficiency: Yancheng Maternity and Child Health Care Hospital reports 2 pathologists now screen 100,000 annual cases with AI assistance, versus 4 pathologists previously required
• High Sensitivity: Maintains detection sensitivity across lesion types including ASCUS, LSIL, and HSIL classifications
• Throughput: System processes digitized slides substantially faster than manual microscopy review

Technical Architecture

The system employs a multi-stage analysis pipeline:

1. Whole-Slide Image Ingestion: Receives digitized LBC slides from KFBIO KF-PRO or KF-PRX scanners
2. Region Detection: Identifies tissue-containing areas and excludes debris or artifacts
3. Cell-Level Analysis: DCNN evaluates individual cells for morphological abnormalities
4. Aggregation and Scoring: Combines cell-level findings into slide-level diagnostic recommendations
5. Report Generation: Outputs structured results with annotated images for pathologist verification

The B/S (Browser/Server) architecture enables convenient access from laboratory workstations while maintaining scalability for high-volume deployments.

Workflow Integration

Screening Program Deployment:

1. LBC specimens prepared using standard protocols
2. Slides digitized on KFBIO high-throughput scanners
3. AI system performs initial screening and classification
4. Negative cases archived with minimal pathologist oversight
5. Suspicious cases queued for detailed manual review
6. Pathologist renders final diagnosis with AI annotations as reference

This workflow addresses regulatory requirements limiting pathologists to 100 manual slide reviews daily while dramatically expanding screening program capacity.

Regulatory Status

The system achieved NMPA registration through standard pathways, with KFBIO pursuing Class III certification via priority review for expanded clinical claims.

Deployment Examples

Yancheng Maternity and Child Health Care Hospital: Integrated AI screening with WSI scanning to enhance cervical cancer prevention capabilities throughout the city, improving both screening throughput and diagnostic quality.

Ningbo Diagnostic Pathology Center: Conducted tens of thousands of controlled clinical trials before deployment, now using AI to screen 100,000+ cases annually with a two-pathologist team.

Brazil National Program: KFBIO contracted to provide cervical cancer screening for hundreds of thousands of women, representing the company’s largest international AI deployment.

Frequently Asked Questions

What percentage of cases does the AI system filter?

Clinical deployments report the AI system filters 50-70% of negative cases from manual review queues, enabling pathologists to focus time on suspicious and positive specimens.

Does the AI system make final diagnoses?

No. The system provides AI-assisted preliminary screening and classification. Licensed pathologists review AI findings and render final diagnostic determinations in compliance with regulatory requirements.

What specimens does the system analyze?

The system is designed for liquid-based cytology (LBC) specimens, the modern standard for cervical cancer screening that has largely replaced conventional Pap smears in developed healthcare systems.

How does the system integrate with existing laboratory workflows?

The AI system receives digitized slides directly from KFBIO scanners and outputs results to the K-Viewer telepathology platform. Laboratories can incorporate AI screening into existing LIS/LIMS workflows through standard integration protocols.