Follow us
Latest News & Events
Pioneering Clinical AI: How a KFBIO Scanner Helped Train a System for Near-Perfect Gastric Cancer Detection
Why High-Quality Digital Pathology Images Matter for AI
The global shortage of pathologists and diagnostic variability continue to challenge gastric cancer diagnosis. Digital pathology and artificial intelligence (AI) are emerging as transformative solutions to these problems.
A landmark study published in Nature Communications demonstrated the potential of this approach. The researchers trained an AI system using images digitized by the KFBIO KF-PRO-005 scanner. The system achieved 99.6% sensitivity and 80.6% specificity in gastric cancer detection.
The KF-PRO-005 also delivers consistent, high-quality whole slide images (WSIs). These images provide a reliable foundation for clinically applicable AI diagnostics.
Introduction: The Diagnostic Challenge and the Digital Solution
Digital pathology scanners are transforming modern pathology workflows.
Gastric cancer remains a major global health burden. Early and accurate diagnosis is essential for improving patient survival. However, healthcare systems worldwide face a growing shortage of pathologists. Heavy workloads and increasing case volumes can also lead to diagnostic variability.
To address these challenges, pathology laboratories are adopting Whole Slide Imaging (WSI). Digital pathology not only modernizes diagnostic workflows but also unlocks the full potential of artificial intelligence.
A landmark study published in Nature Communications highlights the powerful combination of reliable digital scanning and advanced AI. Researchers at the Chinese PLA General Hospital developed a clinically applicable AI system for gastric cancer detection. The system delivered remarkable diagnostic accuracy.
At the center of this pioneering research was the KFBIO KF-PRO-005 digital slide scanner. Researchers selected the scanner to digitize the high-quality training images that enabled the AI to learn from expert pathologists.
Building the AI’s Foundation: Precision Scanning and Expert Annotation
Developing a reliable AI model requires a large, high-quality, and carefully annotated dataset.
For this study, pathologists collected 2,123 H&E-stained whole slide images from 1,500 patients. The dataset included a wide range of tissue types and tumor subtypes. Researchers digitized every slide at 40× magnification (0.238 μm/pixel) using the KFBIO KF-PRO-005 scanner.
Next, a team of 12 senior pathologists performed detailed pixel-level annotations with a custom iPad-based system. They carefully labeled malignant, benign, and other tissue regions. These high-quality annotations, created from consistently scanned images, became the ground truth for training the deep learning model.
Technical Excellence: A Deep Learning Model for Real-World Use
The research team developed a sophisticated deep convolutional neural network based on DeepLab v3 for semantic segmentation.
Unlike conventional classification models, this approach does more than classify a slide as positive or negative. It also generates a detailed pixel-level heatmap that highlights suspicious regions directly on the digital slide. As a result, pathologists can identify potential cancer areas more efficiently.
The researchers also focused on making the model robust enough for routine clinical practice. They introduced advanced training techniques, including color jittering and blur simulation. These methods enabled the AI to perform consistently despite variations in tissue staining and image quality. The results further demonstrate the importance of starting with stable, high-quality whole slide images generated by the KF-PRO-005 scanner.
Breakthrough Performance: Validating the AI Assistant
The research team rigorously evaluated the AI system using a large, real-world test set of 3,212 daily gastric whole slide images (WSIs). The results demonstrated outstanding diagnostic performance.
Sensitivity: 99.6%
The AI correctly identified nearly every malignant case. This exceptionally high sensitivity greatly reduced the risk of missed diagnoses.
Specificity: 80.6%
The system also maintained a high specificity. It correctly identified most benign tissues while supporting efficient clinical workflows.
Robust Performance Across Multiple Scanners
Although researchers trained the model with images from the KFBIO KF-PRO-005 scanner, it also performed reliably on slides digitized by two other leading scanner brands. Notably, the model achieved its highest performance on WSIs generated by the KF-PRO-005, the same platform used during training.
Performance Advantage and Robustness
Reliable performance across different scanning platforms is essential for clinical AI deployment. To evaluate this capability, the researchers tested the model with slides digitized by three scanner models:
- KFBIO KF-PRO-005: 403 WSIs
- Ventana DP200: 977 WSIs
- Hamamatsu NanoZoomer S360: 1,832 WSIs
The AI demonstrated excellent overall stability throughout the evaluation. Across all scanners and testing periods, it achieved an average Area Under the Curve (AUC) of 0.986.
The KF-PRO-005 consistently produced the strongest results because it served as the model’s training platform. Researchers observed only slight performance differences on slides scanned by the other two systems. For example, specificity varied modestly between scanner brands.
These findings highlight the value of using a consistent, high-quality scanning platform throughout AI development and clinical deployment. High-quality image acquisition helps maximize model performance while supporting reliable diagnostic outcomes.
Overall, the results validate two important strengths. First, they demonstrate the excellent image quality delivered by the KF-PRO-005 scanner. Second, they confirm the strong generalizability of the AI model trained with KFBIO-generated images.
a Deep learning model training and inference. We trained the model using WSIs digitalized and annotated at PLAGH. We illustrated the training data distribution at the slide level. The abbreviations are detailed in Supplementary Table 1. The trained model was tested by slides collected from PLAGH and two other hospitals. b The plot of the model performance histogram of the slides from the daily gastric dataset. c Model performance histogram of the daily gastric slides digitalized by three different scanners.
Clinical Impact: Augmenting Pathologists’ Expertise
The study demonstrated more than excellent performance metrics. It also showed how AI can improve daily clinical practice.
Catching Missed Cases
The AI successfully identified two subtle malignant cases that pathologists overlooked during the initial diagnosis. These results demonstrate the system’s value as an additional safety layer for routine pathology practice.
Supporting Challenging Diagnoses
Some complex cases required additional immunohistochemical (IHC) staining before reaching a final diagnosis. In these situations, the AI generated objective probability scores that helped pathologists prioritize, review, and assess difficult cases more efficiently.
Improving Diagnostic Accuracy
The researchers also conducted a timed diagnostic assessment involving 12 junior pathologists. Participants who used the AI-assisted system achieved higher diagnostic accuracy than those who relied only on conventional microscopy or standard digital slides.
These findings demonstrate that AI can complement, rather than replace, pathologists. By combining expert knowledge with intelligent decision support, digital pathology can deliver faster, more accurate, and more consistent diagnoses.
Conclusion: A Step Toward the Future of Pathology
The Nature Communications study provides compelling evidence that AI-powered diagnostic assistance is no longer a theoretical concept. Instead, it has become a clinically viable solution for supporting gastric cancer diagnosis.
The research also highlights the critical role of high-quality digital pathology. Reliable whole slide imaging provides the foundation for accurate AI model development and clinical deployment.
Throughout the study, the KFBIO KF-PRO-005 digital slide scanner delivered the consistent, high-resolution image data required for AI training. These high-quality images enabled researchers to build a trustworthy and highly accurate deep learning model.
Ultimately, this work demonstrates how pathologists and AI can work together. By combining human expertise with machine intelligence, digital pathology can deliver faster, more accurate, and more consistent diagnoses for patients worldwide.
Research Foundation: The application and performance data described herein are based on a landmark clinical study published in Nature Communications. The research, conducted at the Chinese PLA General Hospital, developed a deep learning system for gastric cancer detection using whole slide images digitized by KFBIO scanners. The model demonstrated a sensitivity of 99.6% and a specificity of 80.6% on a large, real-world test set. For full methodological details and results, refer to the original paper: Clinically applicable histopathological diagnosis system for gastric cancer detection using deep learning. Nat Commun11, 4294 (2020).
The KF-PRO-005: The Trusted Foundation for Digital Pathology
Many leading pathology laboratories and research institutions rely on the KFBIO KF-PRO-005 to support digital pathology workflows. The landmark AI study described above is one example of its proven performance.
As one of KFBIO’s flagship digital pathology scanners, the KF-PRO-005 has earned the trust of laboratories worldwide for more than a decade. Today, more than 1,500 units have been installed globally.
The scanner is well known for its stability, reliability, and continuous high-volume operation. It consistently delivers high-quality whole slide images while maintaining excellent scanning performance over time.
In addition, the KF-PRO-005 supports multiple slide sizes and specimen types. This flexibility allows pathology laboratories to meet diverse clinical and research requirements.
Choosing the KF-PRO-005 means investing in more than a digital slide scanner. Laboratories gain a proven platform that supports today’s digital pathology workflow while preparing for future innovations in AI and computational pathology.
Frequently Asked Questions
What role did the KFBIO KF-PRO-005 play in the AI gastric cancer detection study?
Researchers used the KF-PRO-005 to digitize high-resolution whole slide images at 40× magnification (0.238 μm/pixel). These high-quality images provided the essential training data for developing the deep learning model.
How did the AI perform across different scanner brands?
The AI demonstrated robust performance on slides scanned by the KFBIO KF-PRO-005, Ventana DP200, and Hamamatsu NanoZoomer S360. It achieved an average AUC of 0.986 across all platforms, with the strongest performance on images generated by the KF-PRO-005.
What is the clinical significance of this research?
The AI functions as an intelligent decision-support tool rather than a replacement for pathologists. It helps reduce missed diagnoses, supports difficult cases with objective probability scores, and improves diagnostic accuracy, particularly for less experienced pathologists.
Reference
Reference:
This article highlights key findings from the following peer-reviewed study:
Song, Z., Zou, S., Zhou, W. et al. Clinically applicable histopathological diagnosis system for gastric cancer detection using deep learning. Nat Commun 11, 4294 (2020). https://doi.org/10.1038/s41467-020-18147-8
