Revolution in cancer diagnosis: AI conquers pathology!

Revolution in cancer diagnosis: AI conquers pathology!

A significant cooperation project between the University Hospital Erlangen (UKER) and the Gravina Hospital in Caltagirone, Italy, aims to integrate artificial intelligence (AI) into clinical diagnostics in pathology. Loud FAU The results were published in the journal Genome Medicine. With over 1.4 million cancer cases in Germany every year, tissue examinations after tumor removal play a crucial role. This is where AI algorithms come into play, which can help pathologists identify cancer types and assess tissue samples.

The use of AI in pathology is currently proving to be limited because many examinations take place using a microscope. However, Gravina Hospital has taken a step into the future and routinely digitizes all tissue sections. This measure improves the availability of digital data and enables the development of a method for automatically integrating AI analysis into the workflows of pathology laboratories.

Digitalization and AI in pathology

During the diagnosis, tissue samples are processed into extremely fine sections and digitized. The analysis is then carried out on the computer monitor, with the AI ​​analysis automatically activated as soon as new scans are received in the laboratory information system (LIS). Pathologists also have the option of requesting “on-demand” analyses. The results are visualized as heatmaps in the LIS, highlighting cancerous regions. The aim of the project is to improve the accuracy of the algorithms and promote the integration of deep learning models into other pathology departments.

Technological innovations are significantly accelerating cancer diagnosis, particularly through the digitization of histopathological sections and the use of deep learning. An analysis highlights the challenges and opportunities associated with these developments. The use of deep neural networks (deep learning) has the potential to improve diagnostic accuracy and reduce the error rate in cancer diagnoses. For example, a two-stage deep learning model for prostate cancer grading is cited as having achieved an accuracy of 0.7 compared to 0.61 for human pathologists.

The role of GNNs in digital pathology

Of particular note is the development of Graph Neural Networks (GNNs), which are proving to be a promising alternative for feature extraction and interpretability in digital pathology. These networks model the relationships between objects and have already found successful applications in various areas, including the prediction of molecular properties. Research has shown that GNNs can achieve 97% accuracy in grading colorectal cancer, highlighting their superiority compared to traditional approaches.

Implementing AI in pathology is not only a technical challenge but also a necessity. Given the increasing number of cancer cases, the decreasing number of pathologists and the increasing number of cases, there is an urgent need for automated interpretation of digitized pathology images. Fraunhofer IKS highlights that AI can quickly combine and analyze large amounts of data, which can lead to individualized therapy and early disease diagnoses.

Overall, it shows that the combination of digitalization and AI not only improves efficiency and accuracy in pathology, but can also lead to innovative approaches in cancer diagnosis.