Table 2.
Applications and performance of AI algorithms in tumor microenvironment diagnosis of colorectal cancer.
| Article | Year | Aim | Neural network type | Neural network model | Data | Biopsies/resection specimens | CRC stage | Sample size | Sensitivity | Specificity | Accuracy | Precision | Recall |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Kather J. N. et al. Predicting survival from colorectal cancer histology slides using deep learning: A retrospective multicenter study | 2019 | Segmentation for tissue classes: adipose tissue, background, debris, lymphocytes, mucus, smooth muscle, normal colon mucosa, cancer-associated stroma, and CRC epithelium | CNN | VGG19, AlexNet, SqueezeNet v. 1.1, GoogLeNet, Resnet50 | CRC tissue slides stained with H&E | – | – | 100 000 image patches from 86 H&E slides, Training (70%), Validation (15%), Testing (15%) | – | – | >0.94 | – | – |
| Kwak M. S. et al. Deep convolutional neural network-based lymph node metastasis prediction for colon cancer using histopathological images | 2021 | Segmentation for tissue classes: normal colon mucosa, stroma, lymphocytes, mucus, adipose tissue, smooth muscle, and colon cancer epithelium | CNN | U-Net | CRC tissue slides stained with H&E | – | I–III | 100 000 image patches, Training (80%), Validation (10%), Testing (10%) | 0.677 | – | ≈0.85 | – | – |
| Lin A. et al. Deep learning analysis of the adipose tissue and the prediction of prognosis in colorectal cancer | 2022 | Segmentation for tissue classes: tumor and non-tumor sections | CNN | VGG19 | CRC tissue slides stained with H&E | – | – | >100 000 image patches | – | – | 0.973 | – | – |
| Gong C. et al. Quantitative characterization of CD8+ T cell clustering and spatial heterogeneity in solid tumors | 2019 | Detection CD8+ T cells and quantify the spatial heterogeneity | – | – | CRC tissue slides stained with CD8 | – | – | 29 slides | – | – | – | 0.881 | 0.742 |
| Jakab A., Patai Á. V., Micsik T. Digital image analysis provides robust tissue microenvironment-based prognosticators in patients with stage I-IV colorectal cancer | 2022 | Evaluation of tumor-stroma ratio | Machine learning–based algorithm | SlideViewer software 2.4 version and its QuantCenter module | CRC tissue slides stained with H&E | Resection specimens | I–IV | 185 slides | 0.671 (tumor epithelium) 0.646 (stroma) |
0.865 (tumor epithelium) 0.783 (stroma) |
0.80 (tumor epithelium) 0.724 (stroma) |
– | – |
| Failmezger H. et al. Computational tumor infiltration phenotypes enable the spatial and genomic analysis of immune infiltration in colorectal cancer | 2021 | Identification of spatial tumor infiltration phenotypes | Lasso regression model | – | CRC tissue slides stained with Ki67, CD3/CD4, CD3/CD8 | Resection specimens | – | 80 slides | – | – | – | – | – |
| Pai R. K. et al. Development and initial validation of a deep learning algorithm to quantify histological features in colorectal carcinoma including tumour budding/poorly differentiated clusters | 2021 | Segmentation for tissue classes: carcinoma, stroma, mucin, necrosis, fat, and smooth muscle | CNN | – | CRC tissue slides stained with H&E | – | I–IV | – | 0.88 | 0.73 | – | – | – |
| Yoo S. Y. et al. Whole-Slide Image Analysis Reveals Quantitative Landscape of Tumor–Immune Microenvironment in Colorectal CancersTIME Analysis via Whole-Slide Histopathologic Images | 2022 | Evaluation of tumor-infiltrating lymphocytes and tumor-stroma ratio | Random forest classifier | – | CRC tissue slides stained with CD3, CD8 | Resection specimens | II, III | Training cohort (260 slides), Discovery cohort (590 patients), Validation cohort (293 patients). | – | – | – | – | – |
| Zhao K. et al. Artificial intelligence quantified tumour-stroma ratio is an independent predictor for overall survival in resectable colorectal cancer | 2020 | Evaluation of tumor-stroma ratio | CNN | VGG19 | CRC tissue slides stained with H&E | Resection specimens | – | Training set (283100 slides), Test set 1 (6300 slides), Test set 2 (22500 slides) | 0.728 | – | 0.9746 | – | – |
Abbreviation: TME – tumor microenvironment, TB – tumor budding, AUC – area under the curve, CNN – convolutional neural network, CRC – colorectal cancer, H&E – hematoxylin and eosin, CK – cytokeratin.