Case Study: Roche achieves clinical fact extraction from pathology reports with John Snow Labs' Spark NLP

Roche improves clinical fact extraction accuracy with John Snow Labs and OCR

Roche, a global healthcare leader in in-vitro diagnostics, faced the challenge of accurately extracting critical clinical facts from unstructured free-text medical notes, such as pathology and radiology reports. This natural language processing was essential for building their clinical decision support product portfolio but required understanding complex clinical terminology and minimizing the need for extensive task-specific annotations. To tackle this, they turned to John Snow Labs and utilized their Spark NLP for Healthcare product.

John Snow Labs implemented a solution using Spark NLP for Healthcare integrated with OCR technology. This approach leveraged deep learning and transfer learning to achieve higher accuracy on medical data and efficiently retain document structures like tables. The result was a significant minimization in the time required to train models and an automated, scalable system for extracting clinical facts. This enabled Roche to successfully build its clinical decision support applications.

Roche