A computer-based automated algorithm for assessing acinar cell loss after experimental pancreatitis

Abstract

The change in exocrine mass is an important parameter to follow in experimental models of pancreatic injury and regeneration. However, at present, the quantitative assessment of exocrine content by histology is tedious and operator-dependent, requiring manual assessment of acinar area on serial pancreatic sections. In this study, we utilized a novel computer-generated learning algorithm to construct an accurate and rapid method of quantifying acinar content. The algorithm works by learning differences in pixel characteristics from input examples provided by human experts. HE-stained pancreatic sections were obtained in mice recovering from a 2-day, hourly caerulein hyperstimulation model of experimental pancreatitis. For training data, a pathologist carefully outlined discrete regions of acinar and non-acinar tissue in 21 sections at various stages of pancreatic injury and recovery (termed the "ground truth"). After the expert defined the ground truth, the computer was able to develop a prediction rule that was then applied to a unique set of high-resolution images in order to validate the process. For baseline, non-injured pancreatic sections, the software demonstrated close agreement with the ground truth in identifying baseline acinar tissue area with only a difference of 1% ± 0.05% (p = 0.21). Within regions of injured tissue, the software reported a difference of 2.5% ± 0.04% in acinar area compared with the pathologist (p = 0.47). Surprisingly, on detailed morphological examination, the discrepancy was primarily because the software outlined acini and excluded inter-acinar and luminal white space with greater precision. The findings suggest that the software will be of great potential benefit to both clinicians and researchers in quantifying pancreatic acinar cell flux in the injured and recovering pancreas.

Figure 1. Flow diagram depicting the training and testing phases of the computer algorithm.

A human expert establishes the “ground truth.” The computer then creates a prediction rule by modeling image characteristics and analyzing pixel neighborhoods. In the testing phase, the computer applies the prediction rule to a new set of images. The computer results are finally validated by the human expert. White and gray boxes refer to the human expert and computer actions, respectively.

Figure 2. A human expert initiates the training phase of computer learning by identifying the “ground truth.”

(A) HE staining of a whole pancreas slice. Regions to be included are outlined in black. (B) The whole image is parsed into smaller patches. (C, D) A pathologist then identifies acinar structures (in red) and excludes non-acini (in blue) from a heterogeneous array of pancreatic tissues in order to construct the ground truth.

Figure 3. Caerulein hyperstimulation causes marked acinar cell loss.

Representative pancreas sections stained with HE isolated from mice (A) at baseline (uninjured) or (B) 3 days after caerulein hyperstimulation pancreatitis. For demonstration, a single acinus in each panel is outlined in red.

Figure 4. The automated computer-learning system can be used to track acinar cell loss and recovery after acute pancreatic injury.

(A) Representative HE pancreas sections at baseline and from several time points after caerulein hyperstimulation, with acinar structures outlined in red. (B) Quantification of acinar area generated by the computer algorithm and two human experts. (n = 6 patches for each time point). The graph shows absolute percent acinar area compared to total pancreatic tissue (left axis) as well as the relative change in acinar tissue as normalized by Day 3 acinar content (right axis). Overall sensitivity  = 88%, specificity  = 86%, accuracy  = 93%, and F₁ score  = 0.86.

Figure 5. The computer-learning system outlines acini more precisely than can be performed manually.

In the representative HE pancreatic sections from Day 1 after caerulein hyperstimulation, (A) the computer outlines acini (in red) more tightly (arrowheads) and is more likely to exclude intra-acini white space (asterisks) than can be performed manually by (B) human experts.