It is universally known that Digital Pathology (DP) is the process by which glass slides are digitized to produce high-resolution images. However, the promise of digital pathology and the paradigm shift from microscopy to digitization is not merely transfer of an image from glass slide to a monitor; or to share, modify or analyze the image but rather the potential to transform the digital image by augmenting the human eye with information/data to perform intelligent interpretations (algorithmic approaches) that cannot be gleaned by the pathologist’s eye. These digitized slides afford the possibility of applying image analysis techniques to DP for applications in detection, segmentation, and classification. In short, the next breakthrough innovation in digital pathology could be a next-gen smart scanner, which is driven by artificial intelligence and deep learning concept specially designed for the ease of pathologists to scan, index and analyze the tissue slides – all at the same time.
Algorithmic approaches driven by AI and ML technology, have shown to be beneficial in many contexts as they have the capacity to not only significantly reduce the laborious and tedious nature of providing accurate quantifications, but to act as a second reader helping to reduce inter-reader variability among pathologists. Thus, it aids the pathologist to not only look at many thousand cells but pixels which will enable him/her to perform precise measurements of histological patterns on a whole slide image in his/her precision medicine arsenal. “On the fly image analysis solution” embedded in AI-enabled digital pathology scanner is one such algorithmic approach; a combination of artificial intelligence (AI) and Machine Learning (ML) technology.
Characteristics of smart scanner:
A scanner with AI & ML based technology provides fast and accurate diagnostics support in improving medical research.
- ‘On the fly’ cell counting functionality Identifies region of interest (epithelial) as the viewer pans the whole slide scanned image. The functionality also computes nuclear/membrane scores for IHC markers in real time as the user pans the image.
- A heat map is generated on the thumbnail indicating the areas traversed by the user with different color codes for easy reference. User can select the Regions of Interest (ROI) to be added to the report based on the heatmap.
- On the fly concept uses AI and ML techniques to identify and quantify the ROI’s. The ML module is trained based on ground truth to identify and classify different ROI’s based on the intensity, morphology and other intuitive features like clustering of cells and arrangement of cells in spatial domain.
- Another ML module identifies and classifies the cells within the ROI into positive and negative cells for nuclear marker and into a graded classification i.e. 1+, 2+, 3+, negative cells for membrane marker. The cells are classified based on the model generated during training using different cell features like size, shape, intensity etc.
- On the fly cell counting module thus effectively uses ML to quantify the cells in real time and the user does not have to specifically select the region of interests and click on analyze button. The analyzed areas are marked on the thumbnail as heatmap for easy reference and report generation.
Advantage of smart scanner:
These systems offer further advantages by integrating the workflow into the operational environment that is hardware agnostic and lets third parties’ plugins thereby making it a good fit. Augmentation that automatically counts cell types enhances standardization and reproducibility will complement the work of a pathologist by providing more accurate and reproducible analysis of key factors eventually improving patient outcomes.
However, the question bothering the pathology community remains as is! As the technology matures, will computer algorithms outperform human pathologists in both accuracy as well as speed and go beyond assistance, replacing human pathologists eventually? The answer is NO. Artificial intelligence (AI) as a “powerful tool” uses the accuracy of machine learning for complex scoring as well as computational tasks and the cognitive power of the pathologist to analyze complex tissue architecture, each performing to their own strengths. Once in the digital realm, artificial intelligence and deep learning technologies in combination with computational pathology techniques with the humongous data generated will help train deep convolutional neural networks which will mimic the human cognitive capabilities. Multilayered convolutional networks will then perform all complex computational tasks and reveal more information to answer questions about a given disease.
Thus, digital pathology will bring this positive change by replacing the microscope, not the pathologists.