DeepLRHE: A Deep Convolutional Neural Network Framework to Evaluate the Risk of Lung Cancer Recurrence and Metastasis From Histopathology Images

Abstract

It is critical for patients who cannot undergo eradicable surgery to predict the risk of lung cancer recurrence and metastasis; therefore, the physicians can design the appropriate adjuvant therapy plan. However, traditional circulating tumor cell (CTC) detection or next-generation sequencing (NGS)-based methods are usually expensive and timeinefficient, which urge the need for more efficient computational models. In this study, we have established a convolutional neural network (CNN) framework called DeepLRHE to predict the recurrence risk of lung cancer by analyzing histopathological images of patients. The steps for using DeepLRHE include automatic tumor region identification, image normalization, biomarker identification, and sample classification. In practice, we used 110 lung cancer samples downloaded from The Cancer Genome Atlas (TCGA) database to train and validate our CNN model and 101 samples as independent test dataset. The area under the receiver operating characteristic (ROC) curve (AUC) for test dataset was 0.79, suggesting a relatively good prediction performance. Our study demonstrates that the features extracted from histopathological images could be well used to predict lung cancer recurrence after surgical resection and help classify patients who should receive additional adjuvant therapy.

DeepRePath: Identifying the Prognostic Features of Early-Stage Lung Adenocarcinoma Using Multi-Scale Pathology Images and Deep Convolutional Neural Networks

“…Previously, various algorithms using machine learning have been developed to analyze pathology images for cancer detection [31], assessment of the histologic growth pattern [32] and PD-L1 status [33], histological subtyping [34], microenvironment analysis [20,21,25,27], and nuclear segmentation [23]. The prognostic value of these analyses has been described in some studies [18][19][20][21][22][23][24][25][26][27]. The performance of our model (specificity = 78%, sensitivity = 74%, accuracy = 77%, hazards ratio = 5.564, and AUC score = 0.76-0.77) was superior or similar to that of the previously described models [18][19][20][21][22][23][24][25][26][27].…”

“…The prognostic value of these analyses has been described in some studies [18][19][20][21][22][23][24][25][26][27]. The performance of our model (specificity = 78%, sensitivity = 74%, accuracy = 77%, hazards ratio = 5.564, and AUC score = 0.76-0.77) was superior or similar to that of the previously described models [18][19][20][21][22][23][24][25][26][27]. However, these studies first elucidated the specific values, including tumor shape and boundary features [19], tumor cell features [18,23,24], and tumor microenvironment [20,21,25,27], which have already been reported as potential prognostic factors.…”

“…Notably, one of the benefits of using artificial intelligence in pathology is the identification of novel features, that is, subvisual features that could be helpful for supporting decisions in patient management [17]. Several related studies on predicting the prognosis of patients with NSCLC have been published [18][19][20][21][22][23][24][25][26][27]. Previous studies used deep learning to evaluate prognosis-associated histopathological variables, such as tumor-infiltrating lymphocytes, necrosis, tumor stroma, and nuclear segmentation [19,20].…”

An Enhanced Histopathology Analysis: An AI-Based System for Multiclass Grading of Oral Squamous Cell Carcinoma and Segmenting of Epithelial and Stromal Tissue

“…CNNs are considered as the most popular deep learning approach in terms of analysing visual imagery. In their study, authors presented that features extracted from histopathological images can be used for the prediction of lung cancer recurrence [ 19 ]. By analysing histopathological images Tabibu et al presented how deep learning algorithms can be used for pan-renal cell carcinoma classification as well as prediction of survival outcome.…”

The Impact of Digital Histopathology Batch Effect on Deep Learning Model Accuracy and Bias

“…Survival outcomes are particularly challenging to develop rigorous models for using histology from TCGA, and model performance may be falsely elevated not only by the disparate outcomes across sites, but also the site level differences in critical factors relevant to survival such as stage and age. Studies demonstrating histologic discrimination of survival and recurrence in glioblastoma 41,43,44 , renal cell cancer 41 , and lung cancer 45 patients from TCGA which lack external validation cohorts may have biased estimates of outcome. Prediction of survival may also suffer from this bias 46 even when correcting for age, stage, and sex, as other factors that vary by site also contribute to outcome, ranging from ethnicity of enrollees, to the treatment available at academic vs community centers.…”