When a tumour is examined in a pathology laboratory, it is often impossible to observe it as a whole. Before microscopic analysis, the tissue must be divided into several fragments, which are then processed and examined individually. While this procedure is routine in clinical practice, it can make it more difficult to assess the tumour in its entirety.
To support pathologists in this task, researchers from the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI) and the Cantonal Institute of Pathology of the Ente Ospedaliero Cantonale (EOC) developed PATH-zle, an innovative software tool that uses Artificial Intelligence to automatically reconstruct a digital macro-section from images of individual histological fragments.
“The software analyses digital images of the fragments and identifies how they fit together, much like solving a puzzle,” explains Gabriele Abbate, researcher at SUPSI. “Within seconds, it proposes the most likely reconstructions, which can then be reviewed and refined through an intuitive interface.”
The project involved the collection and analysis of 34 real clinical cases, covering different tumour types and a varying number of fragments, from two to ten per sample. In total, researchers worked with 101 ultra-high-resolution digital images.
The results proved highly promising. In 83% of cases, the software identified the correct reconstruction as its first choice. When considering the top three proposed solutions, the success rate increased to 94%. These achievements were also recognised by Innosuisse, which noted that the project exceeded its original objectives both in the size of the dataset collected and in the performance of the developed algorithm.
“Despite significant advances in diagnostic techniques, traditional parameters such as tumour size and the distance between the tumour and the surgical resection margins remain essential for determining prognosis and guiding treatment decisions,” says Prof. Dr. med. Luca Mazzucchelli, Deputy Chief Physician and former Medical and Scientific Director of the Cantonal Institute of Pathology at EOC. “Yet these measurements often require complex and time-consuming procedures that can be difficult to standardise. By automatically reconstructing fragmented tissue sections, PATH-zle provides pathologists with a more complete view of the tumour, supporting faster and more reliable evaluations. This approach has the potential to significantly advance digital pathology and facilitate its adoption in routine clinical practice.”
Beyond generating a reconstruction in less than five seconds, the software produces a final image at the original resolution, fully compatible with the digital platforms already used in pathology institutes. The technology opens new perspectives for digital pathology. Viewing a tumour as a single reconstructed image can facilitate the evaluation of disease extent, surgical margins and responses to cancer treatments.
The prototype is currently being integrated into the clinical workflow of the Cantonal Institute of Pathology of the EOC, while the scientific results are being prepared for publication.