Pathology based analyses are critical in multiple steps of preclinical drug development and clinical trials, as well as regulatory submissions and is a key method to determine mechanisms of disease development as well as therapeutic efficacy and toxicity. Pathology can be broadly divided into 3 types – anatomic, molecular and clinical1. A majority of pathology analysis is performed on both normal and disease tissue samples that are collected, fixed, sectioned onto slides and stained with different dyes or antibodies targeting specific markers. Traditionally, the stained slides are analyzed under a microscope and trained pathologists report the findings on each section. This approach is not scalable and requires a lot of trained human resources so digital pathology methods are being used to improve the speed, scalability and accuracy of the analyses.
Fundamentally, digital pathology covers the acquisition, interpretation and sharing of slide images and data. Analysis starts with whole slide imaging (WSI) where the complete stained section is captured as a digital image that can be shared and viewed on a screen and analyzed using specific algorithms. WSI was approved by the FDA in 2017, and the first WSI platform was Philips IntelliSite Pathology Solution (PIPS)2. WSI preserves the tissue morphology and subcellular structures allowing for detailed analysis on a screen3. and is essential for telepathology where the pathologist is at a different location from the lab. Additionally, digitizing whole sections helps manage the storage requirements for numerous glass slides. Once the digitized slide image is created, it can be stored locally or in the cloud with no loss of quality in contrast to stained slides that can deteriorate over time. Manual analysis by a pathologist is the traditional approach, but increasingly artificial intelligence (AI) based algorithms are being developed to analyze digital slide sections and serve as an aid for pathologists.
Deep learning networks and sophisticated algorithms are currently being used to identify tissue types or differentiate between tumor and normal tissues4. AI-based analysis is increasingly being used in preclinical toxicology studies to improve workflow efficiencies while improving accuracy by reducing variability due to human error5. Image analysis algorithms are useful to differentiate normal and abnormal tissue samples thus allowing pathologists to efficiently analyze large numbers of toxicology samples. Tox studies result in large numbers of samples and datasets, so AI methods like neural networks, large language models (LLM) and natural language processing (NLP) are ideally suited to analyze multiple large datasets to identify patterns and generate actionable insights6. While the potential for AI-based pathology analysis for toxicology studies is very promising, a lot of development and validation is required for AI to be the sole method for pathology analysis. However, these methods are being used to automate specific activities and perform initial analysis which helps improve consistency, accuracy, turn around time and costs for a toxicology program.
Apart from advances in AI methods, researchers are working on novel methods to eliminate tissue sectioning and staining which can take several weeks. Fluorescence imitating brightfield imaging (FIBI) is a new microscopy-based method that uses specific absorbable stains including hematoxylin and eosin (H&E) on thick tissue sections that are then imaged using epifluorescence light7. The fluorescent light generates images of the tissue morphology that can be further analyzed manually or using AI methods7. While this method is in the early stages of development, initial comparison data between standard H&E and FIBI shows a concordance of 97%7,8 suggesting that additional development and validation is warranted.
References:
1https://www.news-medical.net/health/Types-of-Pathology.aspx
3https://pmc.ncbi.nlm.nih.gov/articles/PMC10547926/
4https://pubmed.ncbi.nlm.nih.gov/33421167/
5https://www.aiforia.com/resource-library/toxicologic-pathology
6https://pubmed.ncbi.nlm.nih.gov/38244040/
