Work

A Prognostic and Predictive Computational Pathology Based Companion Diagnostic Approach: Precision Medicine for Lung Cancer

Year: 2019
Type: article
Study design: Observational Study
Abstract: Chapter 1. Introduction 1 OverviewList of Tables 3.1 Configuration of CNN for nuclei segmentation 23 4.1 Demographic and clinical characteristics of patients in cohort #1 and cohort #2 40 4.2 Statistical significance test by Fisher'exact test between gender, major pathological characteristics and disease outcome of 119 NSCLC patients in cohort #2 set.Tow-sided P < 0.05 was considered as statistically significant.41 4.3 A subset of 7 features selected from entire feature set 41 4.4 Classification results with real patient 5-year outcomes of validation set cohort #2.42 4.5 Multivariable Cox proportional hazard model controlling for major pathological variables.P-values in bold are statistically significant 43 4.6 Summary of clinical and pathological features of the studied NSCLC cohorts 51 4.7 List of the most discriminating cluster density features for predicting recurrence in NSCLC.In this study, density is defined as the ratio of the number of cells within the cluster to the cluster pixel area 54 4.8 Comparison of the agreement among experts (Pathologist 1 and 2) and the automated strategy.For comparing manual and automatic strategies, it was assumed that an image graded by a pathologist as "highly infiltrated" has a good prognosis (non-recurrence) 55 xi 4.9 Statistical significance test by Fisher's exact test between gender, major pathological characteristics and disease outcome of patients in Cohort #2 and Cohort #3 set.57 4.10 Multivariate disease specific survival analysis on the test set.58 5.1 Patients' demographics and characteristics 67 5.2 Univariate and Multivariate analysis on validation set (D2 and D3) 72 6.1 Demographic and characteristics of the patients, categorized according to responders and non-responders.86 6.2 Univariate and Multivariate Analysis with OS 89 6.3 Univariate and Multivariate Analysis with TTP 89 xiiList of Figures 1.1 Overall flowchart of the dissertation.Abbreviations: NSCLC: non-small cell lung cancer.2 1.2 Four specific aims for this dissertation.Aim 1 develops the nuclei based QH features and interplay features between cancer and immune cells.Aims 2 applies the subset of Aim 1 developed features to prognosticate survival for early stage NSCLC.Aim 3 constructs a prognostic and predictive model of added benefit of ACT in early stage NSCLC.Aim 4 builds a predictive model for advanced stage NSCLC who respond to ICI based treatment Abbreviations: QH, quantitative histomorphometric; NSCLC, non-small cell lung cancer; TIL, tumor infiltrating lymphocytes; ACT, adjuvant chemotherapy.8 2.1 Examples of image features developed in this dissertation: (a) nuclei segmentation and shape features; (b) nuclei and peri-nuclei image texture; (c, d) nuclei graph; (e) lymphocyte density in local cluster and variance; and (f, g, h) spatial interaction between lymphocyte and cancer nuclei.11 3.1 Generative adversarial U-Net nuclei segmentation model.24 3.2 Example of (a) undirected graph and (b) directed graph.V represents vertex and E represents edge connected vertices.25 3.3 Extraction of density and topological information from TMA cores.a) Original image.b) A watershed-based algorithm is used to segment xiii nuclei in the image.c) Nuclei are classified as lymphocytes (green) and non-lymphocytes (red) based on texture, shape, and color features.Non-recurrence (second row) and recurrence (third row) with corresponding features maps: (d, g) cluster construction, (e, h) cluster area heatmap, and (f, i) cluster density heatmap.29 4.1 Inclusion and exclusion criteria for patient selection for the training and test sets.37 4.2 Flowchart illustrating the procedure for training and validating the quantitative histomorphometric classifier for distinguishing early versus no/late recurrence in early stage lung cancer 38 4.3 Representative TMA histospots of recurrent and non-recurrent NSCLC with corresponding feature maps: Recurrence TMA with (a) nuclear shape feature, (b) texture feature map (Haralick standard deviation intensity correlation), (c) nuclear cluster graph feature map, and (d) nuclear orient.42 4.4 (a) ROC analysis of classifier predicting on training set; (b) ROC analysis of classifier predicting on independent validation set.And Kaplan-Meier survival analysis for (c) training set and (d) validation set show good visual separation and log-rank test indicates the two groups were statistically different (p-value « 0.05).43 4.5 Patients selection workflow for training and validation sets.50 4.6 ROC curve for correctly predicting recurrence in NSCLC using density and topological TIL features.55 xiv 4.7 Kaplan-Meier plots for pathologist's evaluation of TILs.X axis shows the survival time in months and Y axis presents the probability of survival.First row shows the Kaplan-Meier disease specific survival curves for Pathologist 1 on the training and testing set.Second row shows the corresponding survival curves for Pathologist 2 on both datasets.56 4.8 Kaplan-Meier plot for the automated strategy.Subfigure (a) shows the survival curves for training test and is presented just for informative purposes.Subfigure (b) shows the survival curves for the set used for testing.The automated strategy yields a statistically significant prediction of results between recurrence and non-recurrence patients.59 5.1 Patient Selection Diagram from 3 different cohorts: D1 with 100 surgery alone patients were used as discovery set; and D2-4 were used as validation set.66 5.2 Flowchart of overall workflow: (a) automated tumor region prediction; (b) image tile extraction from tumor and (c) nuclei and peri-nuclei features extraction; (d) feature selection and CoRiS calculation; (e) prognostic and predictive validation of CoRiS.71 5.3 Kaplan-Meier plots showing prognostic effect of CoRiS with (a-c) patients' overall survival in D1-3 and (d-f) disease free survival in D1-3.72 xv 5.4 Kaplan-Meier plots showing predictive effect in CoRiS defined different risk of overall survival groups: (a) patients received adjuvant chemotherapy in D2; (b) patients only received surgery in D2; (c) patients received adjuvant chemotherapy in D3; (d) patients only received surgery in D3.Forest plots of different CoRiS defined risk of overall survival groups in (e) D2 and (f) D3. 74 6.1 Patient selection diagram from 3 different cohorts.81 6.2 Flowchart of high-level workflow: A: Expert annotation of tumor B: Tiles processing from annotation C, D: epithelium and stroma separation and E, F: feature extraction.G: feature analysis and selection.H: predictive model construction and validation.85 6.3 Correlation matrix of the two groups of image features for predicting OS and TTP. Green represents the 7 features for predicting the OS and Blue represents the 12 features for predicting TTP.Feature a-s corresponds to features listed in in supplementary table S2-3 accordingly.87 6.4 The Kaplan-Meier survival analysis of OS and TTP in (a, d) D1, (b, e) D2, and (c, f) D3 in pathomics defined low and high-risk groups.88 6.5 ROC analysis of using pathomics features to predict RECIST with features predictive of (a) OS and (b) TTP.88 xvi
Author: Wang, Xiangxue
Language: English
Open Access status: green
OpenAlex ID: W7024265345
FWCI:
Cites: 118
Cited by:
Related to: 10