Automated Nuclear Morphometry: A Deep Learning Approach for Prognostication in Canine Pulmonary Carcinoma to Enhance...

Area of Science:

• Veterinary Pathology
• Computational Pathology
• Artificial Intelligence in Diagnostics

Background:

• Deep learning tools are increasingly integrated into diagnostic workflows for efficiency and reproducibility.
• Nuclear pleomorphism (NP) is a key criterion for malignancy in canine pulmonary carcinoma (cPC) grading.
• Accurate assessment of NP is crucial for prognosis but can be subjective and time-consuming.

Purpose of the Study:

• To evaluate the utility of automated nuclear morphometry using deep learning for assessing NP in cPC.
• To determine the prognostic implications of AI-driven NP assessment in cPC.
• To compare the performance of the deep learning algorithm with manual morphometry and conventional prognostic tests.

Main Methods:

• Development of a deep learning-based algorithm for nuclear segmentation and NP evaluation (anisokaryosis/shape).
• Evaluation on 46 cPC cases with comprehensive follow-up data.
• Comparison of algorithm's NP assessment against manual morphometry, pathologists' estimates, mitotic count, histological grading, and TNM-stage.

Main Results:

• The standard deviation of nuclear area (anisokaryosis) assessed by the algorithm showed good discriminatory ability for tumor-specific survival (AUC=0.80, HR=3.38).
• The algorithm's performance was comparable to manual morphometry.
• Pathologists' NP estimates showed poor inter-observer reproducibility (k=0.204) and variable prognostic value (HR 0.86-34.8); conventional tests lacked significant prognostic value.

Conclusions:

• Fully automated nuclear morphometry using deep learning offers a time-efficient, reproducible, and prognostically valuable method for assessing NP in cPC.
• The AI algorithm demonstrates potential for improving prognostic accuracy in canine lung cancer.
• Further algorithm refinement and validation on larger cohorts are warranted.