Development and initial validation of a deep learning algorithm to quantify histologic features in colorectal carcinoma including tumour budding/poorly differentiated clusters.
Researchers
Journal
Modalities
Models
Abstract
To develop and validate a deep learning algorithm to quantify a broad spectrum of histologic features in colorectal carcinoma.
A deep learning algorithm was trained on hematoxylin & eosin slides from tissue microarrays of colorectal carcinomas (N=230) to segment colorectal carcinoma digitized images into thirteen regions and one object. The segmentation algorithm demonstrated moderate to almost perfect agreement with interpretations by gastrointestinal pathologists and was applied to an independent test cohort of digitized whole slides of colorectal carcinoma (N=136). The algorithm correctly classified mucinous and high-grade tumours and identified significant differences between mismatch repair proficient and deficient (MMRD) tumours with regards to mucin, inflammatory stroma, and tumor infiltrating lymphocytes (TILs). A cutoff of >44.4 TILs per mm2 carcinoma demonstrated a sensitivity of 88% and specificity of 73% in classifying MMRD carcinomas. Algorithm measures of tumour budding and poorly differentiated clusters (TB/PDC) outperformed TB grade derived from routine sign-out and compared favorably to manual count of TB/PDC with regards to lymphatic, venous, and perineural invasion. Comparable associations were seen between algorithm measures of TB/PDC and manual count of TB/PDC for lymph node metastasis (all P<0.001); however, stronger correlations were seen between the proportion of positive lymph nodes and algorithm measures of TB/PDC. Stronger associations were also seen between distant metastasis and algorithm measures of TB/PDC (P=0.004) compared with TB (P=0.04) and TB/PDC counts (P=0.06).
Our results highlight the potential of deep learning to identify and quantify a broad spectrum of histologic features in colorectal carcinoma.
This article is protected by copyright. All rights reserved.