In response to the global shortage of pathologists and diagnostic variability in gastric cancer, digital pathology and AI have emerged as transformative solutions. Notably, a landmark study in Nature Communications demonstrates that an AI system trained on images digitized by the KFBIO KF-PRO-005 scanner achieved 99.6% sensitivity and 80.6% specificity in gastric cancer detection. Also, the KF-PRO-005 provides consistent, high-quality whole slide images. Therefore, it forms a reliable foundation for clinically applicable AI diagnostics.

Introduction: The Diagnostic Challenge and the Digital Solution
One innovative tool transforming this field is the digital pathology scanner.

Gastric cancer remains a significant global health burden, where early and accurate diagnosis is critical for patient survival. However, this mission is challenged by a worldwide shortage of pathologists. In turn, heavy workloads and potential diagnostic variability arise as a result. The digitization of pathology through Whole Slide Imaging (WSI) has emerged as a transformative solution. In addition, it modernizes the workflow and unlocks the potential of Artificial Intelligence (AI).

A landmark study published in Nature Communications demonstrates the powerful synergy between reliable digital scanning and advanced AI. Conducted at the Chinese PLA General Hospital, the research developed a clinically applicable AI system for detecting gastric cancer. This system provided remarkable accuracy. At the foundation of this pioneering research was the KFBIO KF-PRO-005 digital slide scanner. It was chosen to digitize the high-quality training images that taught the AI to see like an expert pathologist.

Building the AI’s Foundation: Precision Scanning and Expert Annotation

The development of a robust AI model requires a massive, high-quality, and meticulously annotated dataset. For this study, pathologists curated 2,123 H&E-stained whole slide images from 1,500 patients. These slides encompassed a diverse range of tissue types and tumor subtypes. In addition, they were digitized at high resolution (40x magnification, 0.238 μm/pixel) using the KF-PRO-005 scanner.

A dedicated team of 12 senior pathologists used a custom iPad-based system to perform pixel-level annotations, meticulously outlining areas as malignant, benign, or otherwise. This “ground truth” data, captured from crisply scanned images, served as the essential textbook for the deep learning model.

Technical Excellence: A Deep Learning Model for Real-World Use

The research team employed a sophisticated deep convolutional neural network (DeepLab v3) designed for semantic segmentation. This approach allows the AI to not just classify a slide as positive or negative. It also lets the AI generate a detailed, pixel-level “heatmap.” As a result, this highlights suspicious regions directly on the digital slide.

A key focus was ensuring the model’s robustness for real-world clinical use. The training incorporated advanced techniques like color jittering and blur simulation to ensure the AI could perform consistently across variations in tissue staining and image quality—a testament to the need for starting with the consistent scan quality provided by scanners like the KF-PRO-005.

Breakthrough Performance: Validating the AI Assistant

The AI system’s performance was rigorously validated on a large, real-world test set of 3,212 daily gastric WSIs. The results were exceptional:

• Sensitivity: 99.6% – The system correctly identified nearly every malignant case, minimizing the risk of missed diagnoses.

• Specificity: 80.6% – It maintained a high rate of correctly identifying benign tissue, ensuring efficiency.

• Robust Performance Across Scanners: While trained on KFBIO scans, the model performed stably on slides digitized by two other major scanner brands, proving its generalizability. Notably, the performance was highest on WSIs from the KF-PRO-005, its native training platform.

Performance Advantage and Robustness
A crucial test for any clinical AI system is its ability to perform reliably across different hardware. In this study, the test slides were digitized by three scanner models. These models were the KFBIO KF-PRO-005 (403 WSIs), Ventana DP200 (977 WSIs), and Hamamatsu NanoZoomer S360 (1,832 WSIs).

The model demonstrated outstanding overall stability, with an average AUC (Area Under the Curve) of 0.986 across all scanners and time periods. Notably, the highest performance was consistently observed on WSIs produced by the KF-PRO-005 scanner, the very same model used for training. In comparison to the KF-PRO-005 baseline, slight performance variations were noted on slides from other scanners. For example, specificity changes were observed. This underscores the advantage of using a consistent, high-quality scanning source like the KF-PRO-005 for both development and deployment. As a result, optimal AI performance is achieved. This result powerfully validates both the superior image quality of the KF-PRO-005 and the robust generalizability of the AI model trained on its data.

Clinical Impact: Augmenting Pathologists’ Expertise

The study went beyond metrics to demonstrate tangible clinical utility:

1. Catching Missed Cases: The AI successfully flagged two subtle malignancies that had been overlooked in initial human diagnoses, showcasing its potential as a critical safety net.

2. Aiding Difficult Diagnoses: For challenging cases requiring additional immunohistochemical (IHC) stains, the AI’s probability scores provided valuable, objective guidance, helping pathologists triage and assess cases.

3. Improving Diagnostic Accuracy: In a timed assessment with 12 junior pathologists, those using the AI assistance system achieved higher diagnostic accuracy compared to groups using only microscopes or plain digital slides.

Conclusion: A Step Toward the Future of Pathology

The Nature Communications study provides compelling evidence that AI-powered diagnostic assistance is not just a theoretical concept but a clinically viable tool. It underscores the critical role of high-fidelity digitization as the first step in this journey. The KFBIO KF-PRO-005 scanner proved to be an indispensable partner in this research, providing the consistent, high-resolution image data necessary to build a trustworthy and powerful AI.

This work paves the way for a future where pathologists and AI systems work in tandem—combining human expertise with machine precision to achieve 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

At the heart of many groundbreaking pathology workflows and research projects—like the landmark AI study featured above—is the KFBIO KF-PRO-005. As a star model in our lineup, it has earned the trust of laboratories worldwide for over a decade. In fact, there have been over 1,500 units sold globally.

Renowned for its exceptional stability and reliability, the KF-PRO-005 is engineered for continuous, high-volume operation. As a result, it ensures consistent scan quality day after day. Its versatility is a key advantage, offering comprehensive support for different slide sizes and types. This meets the diverse needs of any pathology department.

By choosing the KF-PRO-005, laboratories don’t just acquire a scanner. They invest in a proven, robust platform that digitalizes their present and unlocks the potential of future innovations in AI and computational pathology.

Q: What role did the KFBIO KF-PRO-005 play in the AI gastric cancer detection study?
A: The scanner was used to digitize high-resolution whole slide images (40×, 0.238 μm/pixel), providing the high-quality training data essential for developing a deep learning model with exceptional diagnostic accuracy.

Q: How did the AI perform across different scanner brands?
A: The model demonstrated robust performance across KFBIO KF-PRO-005, Ventana DP200, and Hamamatsu NanoZoomer S360 scanners (average AUC 0.986), with the highest consistency and accuracy observed on KF-PRO-005 digitized slides.

Q: What is the clinical significance of this research?
A: The AI system acts as an assistive tool, reducing missed diagnoses (it identified initially overlooked malignancies), providing objective guidance for challenging cases, and improving diagnostic accuracy among junior pathologists.

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