Abstract
Mechanistic disease progression studies using animal models require objective and quantifiable assessment of tissue pathology. Currently quantification relies heavily on staining methods which can be expensive, labor/time-intensive, inconsistent across laboratories and batch, and produce uneven staining that is prone to misinterpretation and investigator bias. We developed an automated semantic segmentation tool utilizing deep learning for rapid and objective quantification of histologic features relying solely on hematoxylin and eosin stained pancreatic tissue sections. The tool segments normal acinar structures, the ductal phenotype of acinar-to-ductal metaplasia (ADM), and dysplasia with Dice coefficients of 0.79, 0.70, and 0.79, respectively. To deal with inaccurate pixelwise manual annotations, prediction accuracy was also evaluated against biological truth using immunostaining mean structural similarity indexes (SSIM) of 0.925 and 0.920 for amylase and pan-keratin respectively. Our tool’s disease area quantifications were correlated to the quantifications of immunostaining markers (DAPI, amylase, and cytokeratins; Spearman correlation score = 0.86, 0.97, and 0.92) in unseen dataset (n = 25). Moreover, our tool distinguishes ADM from dysplasia, which are not reliably distinguished with immunostaining, and demonstrates generalizability across murine cohorts with pancreatic disease. We quantified the changes in histologic feature abundance for murine cohorts with oncogenic Kras-driven disease, and the predictions fit biological expectations, showing stromal expansion, a reduction of normal acinar tissue, and an increase in both ADM and dysplasia as disease progresses. Our tool promises to accelerate and improve the quantification of pancreatic disease in animal studies and become a unifying quantification tool across laboratories.
Highlights
Advances in deep learning technologies are creating opportunities for the rapid and objective assessment of both normal tissue and pathologic processes in biologic specimens
The earliest stages of oncogene-induced pre-cancer evolution are marked by an expansion of ductal cells or by the conversion of the acinar cells to a ductal phenotype in an adaptive process known as acinar-to ductal metaplasia (ADM)[13]
In order to predict the histologic feature distributions and immunofluorescent stain positivity in murine pancreatic pre-cancerous tissues, several UNet m odels[19] were trained using intensity normalized Hematoxylin and eosin (H&E) image tiles paired with annotated ground truth tiles (Supplemental Fig. 1)
Summary
Advances in deep learning technologies are creating opportunities for the rapid and objective assessment of both normal tissue and pathologic processes in biologic specimens. The extensive and growing use of murine models in disease studies creates a significant need for tissue assessment methods that are rapid, objective and quantifiable in order to permit statistically validated disease measurements among animal cohorts, free of technical variability and investigator bias. Using murine models of pancreatic cancer progression and pancreatitis, we are working to develop and validate deep learning approaches that enable the rapid, reliable, and automated quantification of disease progression over large tissue areas, solely based on H&E staining. Low-grade dysplasia is a pre-invasive neoplasia that can evolve to high-grade dysplasia (PanIN 3) and progress to invasive pancreatic ductal adenocarcinoma (PDAC)[14] Both ADM and dysplasia are accompanied by a prominent stromal reaction and immune cell infiltrate[13]. The stages of ADM and dysplasia evolution are believed to encompass a long phase of pre-cancer evolution that is a valuable window for early intervention[14]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
