AI Model Excels at Analyzing Diverse Cancer Types and Unseen IHC Data

Despite the existence of several AI algorithms to assist in scoring IHC images and improving diagnostic accuracy, current AI models face significant challenges, including data dependency and a lack of generalization. These AI-IHC models require large datasets of immunostain-specific images for training, which are often difficult to obtain, especially for newly developed immunostain-target pairs. Furthermore, these models struggle to analyze datasets that differ from their training set in terms of immunostain or cancer types, limiting their effectiveness across diverse clinical indications. These limitations highlight the need for scalable AI solutions capable of providing accurate analysis across a broad range of cancer types and immunostains. A new study has now demonstrated how an artificial intelligence (AI) model can excel at analyzing diverse cancer types and IHC stains, including datasets it had never previously encountered, due to an innovative training approach.

Lunit (Seoul, South Korea) has developed the Universal Immunohistochemistry (uIHC) AI model, now commercialized as Lunit SCOPE uIHC, which enables advanced biomarker analysis from even singleplex IHC, including subcellular stain localization, continuous intensity scoring, and cell type identification. In a study, Lunit compared eight deep learning models, including four single-cohort models (trained using data from a single stain or cancer type) and four multi-cohort models (trained on datasets that span multiple stains and cancer types), to assess their performance on both familiar and unseen datasets. The results, published in npj Precision Oncology, demonstrated that the uIHC model can generalize across diverse datasets with high accuracy.

The findings underscore the model's strong performance across a wide array of cancer types and immunostains, including those it had not been trained on. The uIHC model’s ability to generalize across different IHC images represents a significant advancement in digital pathology. By reducing the need for large, stain-specific datasets, this model facilitates scalable and efficient biomarker analysis, which is crucial for clinical diagnostics and drug development. This capability is particularly beneficial in evaluating new biomarkers related to emerging therapies, helping to address a major bottleneck in precision oncology.

"Our Universal Immunohistochemistry AI model solves a practical bottleneck in development settings—handling unseen cancer types and stains without requiring additional data annotation," said Brandon Suh, CEO of Lunit. "By proving the effectiveness of a multi-cohort training approach, this study shows how AI can be adapted to real-world complexities, delivering both precision and scalability. With the launch of Lunit SCOPE uIHC, we're enabling researchers and clinicians to focus on what truly matters: advancing patient care and accelerating therapeutic innovation."

Related Links:
Lunit