Artificial Intelligence – Influencing The Clinical Practice Of Pathologists
Posted 3rd January 2018 by Jane Williams
Artificial intelligence based applications are quickly improving and they are expected to be a game changer in many fields, including transportation and medicine. How artificial intelligence will be capable to positively influence the clinical practice of pathologists is a controversial topic that has both supporters and opponents. Digital pathology image retrieval is an intermediate solution that allows pathologists to save time and enhance their clinical performance while letting them control the artificial intelligence resources according to their experience and competence.
Image Retrieval Systems
In the last 20 years, we have benefited from information retrieval in our everyday life due to web search engines which allow us to browse the internet. We can access, with unprecedented rapidity, information starting from just a few concepts and words.
An image retrieval system is a system for browsing, searching and retrieving images from a large database of digital images. Currently, web search engines often include image search systems, allowing us to search for images that result similar to a specific concept or to an image provided by the user. Image retrieval systems are based on two main techniques: text search and visual features (textures, colours and shapes).
Clinical practices are getting increasingly fatiguing for pathologists due to increasing complexity and time constraints. Cancer grading is often complex. The detailed analysis of a single case can require several slides with multiple stainings and quantitative parameters (mitoses counting) are increasingly required.
Digital Pathology and Histopathology
Histopathology is slowly shifting to digital pathology which is opening unprecedented opportunities to save time. The pathologist’s visual inspection of tissue samples is currently the standard method to diagnose a considerable number of diseases and to grade/stage most types of cancer (Gurcan et al., 2009; Rubin et al., 2008). Digital pathology was first applied in the 1980s, but it started to be used in clinical practice relatively recently. This is due to several factors such as slow scanning speed, poor quality on screen, high costs, required memory and limited network bandwidth.
The first high-resolution, automated, Whole-Slide Imaging (WSI) system was developed in 1999 (Ho et al., 2006). Since then, the interest in WSI for pathology has continuously grown because it overcomes the limitations of previous image acquisition methods such as poor image quality and image navigation (Weinstein et al., 2009). After whole slide imaging, the shift to a fully digital environment can be currently expected for pathology, just as it previously happened for radiology. Digital pathology holds tremendous opportunities for histopathology practice. For instance, it can allow online consultations, it can provide access to pathology services in remote areas and it allows applying computer vision and artificial intelligence algorithms to histopathology.
Image Retrieval in Digital Pathology
Digital pathology can apply image retrieval for histopathology as well. Several applications have been performed in the past, working mainly in scientific environments (Caicedo et al., 2014; Doyle et al., 2007; Zhang et al., 2014). Recently, multimodal data fusion was applied to pathology reports and text annotations on prostate frozen sections, cervical cancer, basal-cells and several other tissue types (Caicedo et al., 2014; Jimenez-del-Toro et al., 2017), showing that it can benefit both text and visual information in histopathology. Moreover, content-based image retrieval was used to localise regions of interest and to retrieve breast cancer images (Qi et al., 2014; Zhang et al., 2015).
However, the retrieval of digital pathology images is not an easy task due to the high variability of staining procedures and the very high variety and complexity of histopathology images.
In our presentation at the Digital Pathology Congress, we introduced an innovative retrieval system for digital pathology. The system is currently used for scientific research purposes, but it will soon be translated to a market product. The system is integrated within a viewer, allowing it to define precisely the region of interest to be retrieved. It is based on a multimodal approach, allowing it to exploit both text information and content-based image features. It can also search similar images on diverse data sources, including proprietary datasets (that should ideally be provided by the owner) and public datasets (including scientific literature).
The histopathology retrieval system can be a useful tool for pathologists. It can allow them to save time, reduce the need for consultations and second opinion requests, while also making learning and remembering easier tasks. It is expected to speed up and enhance the clinical practice of pathologists and to make them capable to use their experience and knowledge to properly control artificial intelligence resources.
Manfredo Atzori is a Senior Researcher at the University of Applied Sciences Western Switzerland. He recently presented at the 4th Digital Pathology Congress Europe on Image Retrieval Systems.
If you enjoyed this article, check out ‘Digital Pathology and Artificial Intelligence in the Precision Medicine Era‘.
Leave a Reply