Integrating AI into Pathological Diagnosis: A Seamless Approach

Advancements in artificial intelligence (AI) are increasingly transforming the landscape of medical diagnostics, particularly in pathology. A collaborative initiative involving Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) and Gravina Hospital in Caltagirone, Italy, illustrates how AI can be effectively incorporated into clinical settings, particularly within a fully digital pathology department. These findings have been documented in the journal Genome Medicine.

In Germany alone, over 1.4 million patients are diagnosed and treated for cancer annually. Following the surgical removal of a tumor, the pathological examination of tissue samples is critical for determining the cancer type, assessing malignancy, and deciding on treatment options such as chemotherapy. AI algorithms offer assistance in these areas, for instance, by identifying malignant transformations in digitized tissue samples. Despite this potential, many healthcare facilities have not fully harnessed AI technologies due to challenges in the examination process.

Conventional pathology practices rely heavily on microscopes to assess tissue samples, which limits the integration of AI. According to a researcher from the Department of Nephropathology at UKER, the digitization of histopathological samples to yield high-resolution images is still not widespread. Conversely, Gravina Hospital has taken significant strides in this area by routinely digitizing all tissue samples, addressing the common issue of data availability.

However, a major hurdle has been the lack of automated analysis capabilities for these digital datasets using deep learning models. The research team sought to develop a more streamlined integration of AI tools into their workflow. Their process begins when a tissue sample is received in the pathology lab, where it undergoes several preparatory steps, including the creation of thin sections mounted on glass slides that are then stained with various chemicals. High-resolution digital images of these slides are generated, which can be accessed through the laboratory information system (LIS).

Instead of relying on traditional microscope evaluations, pathologists can now make diagnoses directly on computer screens. The researchers have successfully created a method that automatically incorporates AI analysis into the diagnostic workflow. When new images are uploaded to the LIS, the relevant data for analysis is automatically sent to a server equipped with various AI models. The appropriate algorithms are then selected based on the staining method and tissue type.

This enhanced integration is expected to improve the accuracy of AI algorithms in pathology. The results of the analysis are subsequently returned to the LIS, where predictions are displayed as heatmaps, visually highlighting malignant areas within the digitized tissue samples. The research team aims to validate the integrated deep learning models clinically, with hopes of enhancing their accuracy and expanding the application of these models to routine diagnostics across other pathology departments.

This collaborative project represents a significant step towards bridging the gap between computational pathology and everyday clinical practices, ultimately aiming for improved diagnostic precision and patient outcomes.