Providing a quantitative assessment of tumour nuclei would improve decision objectivity and overcome inter and intra-observer variation. In this study, we show that the summary statistics for the whole slide image of nuclear pleomorphism can provide such quantification. We characterise the heterogeneity of lung adenocarcinoma (LUAD) using morphometric features of tumour nuclei. The Cox proportional hazard regression model is employed on a dataset of 78 patients to find the top discriminative

more »

... tures such that there is a strong correlation with patient survival. We find that global nuclear morphometric features, characterised by heatmap statistics, have a significant correlation with overall survival in LUAD (p < 0.0003). INDEX TERMS Digital pathology, deep learning, lung adenocarcinoma, whole slide image. 12322 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ VOLUME 9, 2021