AIVIS Presents Two Studies Advancing Pathology AI for Real-World Clinical Use at International Conferences
etnews
November 13, 2026

Validating technology that stabilizes AI analysis performance across multi-scanner environments as Qanti® IHC deployments expand | Extending IHC-based cell analysis to H&E, demonstrating the multi-biomarker potential of Qanti® Discovery AIVIS (CEO Daehong Lee), a company specializing in AI-powered digital pathology and precision biomarker analysis, announced that it presented two research studies at international conferences in the United States and Singapore, focused on developing biomarker analysis AI that performs reliably in real-world clinical settings and on expanding the scope of its analysis. At Pathology Visions 2026, held in San Diego, USA, from October 16 to 18, and ASDP 2026, held in Singapore from November 1 to 3, AIVIS unveiled research on stabilizing AI analysis performance across multi-scanner environments and on applying immunohistochemistry (IHC) data to H&E analysis, respectively. Beyond simply improving AI model performance or developing new training techniques, the two studies are significant in that they secure the technical stability required as hospital deployments expand, and extend the company's accumulated pathology analysis technology into new testing and biomarker analysis areas. AIVIS supplies Qanti® IHC, an AI-based IHC image analysis solution approved by the Ministry of Food and Drug Safety (MFDS), to medical institutions in Korea. As its on-site footprint—including paid deployments, adoptions, and proof-of-concept projects—has grown to around 20 medical institutions, maintaining stable analysis performance across the different digital pathology equipment and imaging environments of each hospital has emerged as a key technical challenge. Medical institutions use pathology slide scanners from different manufacturers such as Philips, Leica, and Hamamatsu. Even when the same tissue is scanned, color, contrast, and sharpness can vary with each device's optical characteristics, which can affect AI cell detection and quantitative analysis. It is therefore essential not only to achieve high accuracy on a specific laboratory device, but also to deliver consistent results across diverse real-world environments. At Pathology Visions 2026, the AIVIS research team presented UniScan-Ki67, a study that reduces the degradation of AI cell detection performance caused by differences between scanners in IHC image analysis. The team developed a normalization technique that uses generative AI to align images captured by different scanners with the image characteristics of a reference scanner. With this technique, the detection performance metric (mAP50) on images from two scanners improved from 0.079 to 0.795 and from 0.350 to 0.799, respectively, without any additional training of the cell detection model—close to the performance on reference scanner images used for model training (0.814). The results experimentally validate that image normalization can substantially reduce performance degradation without retraining the core cell detection model each time the scanner changes. AIVIS continues to advance its multi-scanner technology, which has become critical in real-world Qanti® IHC deployments, and this study quantitatively evaluated the technology in Ki-67 analysis. At ASDP 2026 in Singapore, AIVIS also presented a study extending the cell analysis data and AI technology it has accumulated on IHC images to H&E image analysis. Conducted jointly by AIVIS and the Department of Pathology at Korea University College of Medicine, the study examined whether cell location and classification information already built on IHC images can be used to train new H&E image analysis models. The team developed an AI model that transfers cell annotations from the original images to H&E through virtual staining, and used it to train a cell detection model. When evaluated on an external dataset of 24 real breast cancer H&E slides comprising 100 analysis regions, the model trained with virtual H&E images improved cell detection recall on real H&E images by about 10 percentage points compared with a model trained on IHC images alone. The study suggests that leveraging cell information already marked on IHC images can reduce the burden of building training data and broaden the application of pathology AI to H&E image analysis. AIVIS is linking these results to the technology roadmap of Qanti® Discovery, its research pathology AI platform that supports cell-level quantitative analysis of IHC-based biomarkers. The company plans to extend its analysis technology to H&E and other tissue images, broadening the analysis areas available for drug development and pathology research. To this end, AIVIS will strengthen the integration of IHC and H&E analysis and develop technology capable of comprehensively analyzing tumor and immune cell distribution, the tumor microenvironment, and the expression characteristics of various IHC biomarkers. By applying its existing cell-level analysis data and technology to new staining and testing areas, the company aims to streamline the development and validation of biomarker analysis models. Daehong Lee, CEO of AIVIS, said, "For pathology AI to create value in real clinical settings, high accuracy on a specific dataset is not enough. It must work reliably across the different equipment and testing environments of each hospital and provide analysis results that clinicians and researchers can trust. AIVIS is incorporating the technical requirements identified through real-world Qanti® IHC deployments into its R&D and validating them quantitatively." He added, "At the same time, we are extending the vast cell-level analysis technology and data accumulated in IHC to H&E and a range of immune biomarker research. This is not simply about developing research AI models, but about building a practical pathology AI foundation that spans the pathology analysis needed by medical institutions as well as biomarker research and drug development at pharmaceutical companies."