Development of a Deep-Learning Pipeline to Recognize and Characterize Macrophages in Colo-Rectal Liver Metastasis

SIMPLE SUMMARY: We recently proved that in human colorectal cancer, the presence of small or large tumor-associated macrophages (TAMs) is associated with different outcomes. To translate this biological data into a robust clinical marker means to identify in a single slide all TAMs, hundreds of cells, and then evaluate the area of each of them, a task unfeasible in the routine pathology workout. With the aim to develop a deep-learning pipeline to tackle this challenge, we selected, trained and tested three different approaches. The deep-learning pipeline based on the DeepLab-v3 architecture and semantic segmentation technique warrants the separation of TAMs from the background and the identification of single TAMs: this will easily allow the evaluation of their area. ABSTRACT: Quantitative analysis of Tumor Microenvironment (TME) provides prognostic and predictive information in several human cancers but, with few exceptions, it is not performed in daily clinical practice since it is extremely time-consuming. We recently showed that the morphology of Tumor Associated Macrophages (TAMs) correlates with outcome in patients with Colo-Rectal Liver Metastases (CLM). However, as for other TME components, recognizing and characterizing hundreds of TAMs in a single histopathological slide is unfeasible. To fasten this process, we explored a deep-learning based solution. We tested three Convolutional Neural Networks (CNNs), namely UNet, SegNet and DeepLab-v3, with three different segmentation strategies, semantic segmentation, pixel penalties and instance segmentation. The different experiments are compared according to the Intersection over Union (IoU), a metric describing the similarity between what CNN predicts as TAM and the ground truth, and the Symmetric Best Dice (SBD), which indicates the ability of CNN to separate different TAMs. UNet and SegNet showed intrinsic limitations in discriminating single TAMs (highest SBD [Formula: see text]), whereas DeepLab-v3 accurately recognized TAMs from the background (IoU [Formula: see text]) and separated different TAMs (SBD [Formula: see text]). This deep-learning pipeline to recognize TAMs in digital slides will allow the characterization of TAM-related metrics in the daily clinical practice, allowing the implementation of prognostic tools.