Virtual Library
Start Your Search
Kaustav Bera
Author of
-
+
MA15 - Usage of Computer and Molecular Analysis in Treatment Selection and Disease Prognostication (ID 141)
- Event: WCLC 2019
- Type: Mini Oral Session
- Track: Pathology
- Presentations: 1
- Now Available
- Moderators:John Le Quesne, Noriko Motoi
- Coordinates: 9/09/2019, 15:45 - 17:15, Tokyo (1982)
-
+
MA15.05 - Computerized Measurements of Cellular Diversity on H&E Tissue Are Prognostic of OS and Associated with Mutational Status in NSCLC (Now Available) (ID 1975)
15:45 - 17:15 | Author(s): Kaustav Bera
- Abstract
- Presentation
Background
Tumor heterogeneity is known to be implicated in chemotherapeutic resistance and poor prognosis for non-small cell lung cancer (NSCLC). In this study we sought to evaluate the role of computer extracted features reflecting the intrinsic cellular morphological diversity (ICMD) of tumors from digitized H&E stained images of early-stage NSCLC patients. Additionally, we sought to evaluate the association of these ICMD features in adenocarcinomas with the ALK and EGFR mutational status.
Method
Two cohorts, D1 and D2, of digitized H&E stained tissue microarray images (TMA) of NSCLC, n=395 and n=91, respectively, were used for modeling the ICMD predictor. A pretrained deep learning model was used for segmentation of nuclei, and clusters of proximally located nuclei were identified. The ICMD features were then extracted as the variations in shape, size, and texture measurements of nuclei within the clusters. A Cox proportional hazard model using the ICMD features was then trained for lung adenocarcinomas (LUAD, n=270), and squamous cell carcinomas (LUSC, n=216), separately, and was validated on independent cohort from (D3) The Cancer Genome Atlas (TCGA) (n=473) to predict Overall Survival (OS). Univariate and multivariate analyses were performed on (D3).
Result
In (D3), high risk patients predicted by the ICMD features had significantly poorer survival (HR (95% CI) = 1.48 (1.06-2.06), p=0.021 for LUSC, HR (95% CI) = 1.59 (1.11-2.29), p=0.006 for LUAD) in univariate analysis. In multivariate analysis, controlling for major clinical variables, ICMD was independently associated with 5-year OS (p<0.016). (See Table 1) We also found that ICMD features were associated with driver mutations ALK (p=0.0204) and EGFR (p=0.0017) in LUAD.
Table 1| Multivariate analysis for overall survival on the validation set D3.
ConclusionMultivariate Cox Proportional Hazard Model Analysis Controlling for Other Variables
TCGA-LUSC
TCGA-LUAD
Variable
HR (95% CI)
p value
HR (95% CI)
p value
Age (>65 vs <=65)
1.14(0.81-1.61)
0.451
0.89(0.63-1.28)
0.540
Smoking status
1.36(0.83-2.23)
0.221
1.14(0.64-2.01)
0.661
Overall Stage (Stage II vs I)
1.13(0.66-1.94)
0.651
1.86(1.04-3.32)
0.037
T-Stage (T2,3 vs T1)
1.26(0.85-1.87)
0.244
1.25(0.85-1.85)
0.263
N-Stage (N1 vs N0)
1.36(0.77-2.41)
0.292
3.11(1.55-6.23)
0.001
Developed Model
High risk vs. Low risk
1.52(1.08-2.13)
0.016
1.55(1.09-2.22)
0.015
CI = 95% confidence interval; HR = Mantel-Haenszel Hazard ratio. Values in bold are statistically significant, p<=0.05.
Computer extracted image features of cellular diversity were able to predict OS in NSCLC and were also associated with the ALK and EGFR mutational status. Future work will entail evaluating ICMD features in predicting added benefit of adjuvant therapy in early stage NSCLCs as well as correlating with gene expression data.
Only Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.
-
+
MA25 - Precision Medicine in Advanced NSCLC (ID 352)
- Event: WCLC 2019
- Type: Mini Oral Session
- Track: Advanced NSCLC
- Presentations: 1
- Now Available
- Moderators:Sebastian Defranchi, Karen Kelly
- Coordinates: 9/10/2019, 14:30 - 16:00, Dublin (1997)
-
+
MA25.02 - Arrangement and Architecture of Tumor-Infiltrating Lymphocyte on H&E Slides Predict OS in Nivolumab Treated Non-Small Cell Lung Cancer (Now Available) (ID 2911)
14:30 - 16:00 | Presenting Author(s): Kaustav Bera
- Abstract
- Presentation
Background
Immune checkpoint inhibitors (ICI) are a promising and novel approach to treating chemotherapy refractory advanced NSCLC as well as first-line combination therapy in certain NSCLC. Nivolumab, a PD-L1 inhibitor is a promising ICI showing durable benefit with low toxicity in these patients. While PD-L1 positivity is an established tissue based biomarker for response to Nivolumab, studies have shown response rates ranging from 20-50%. Recent research has shown that TILs have been implicated in cancer aggressiveness as well as immune response. In this work, we go beyond simply counting TILs, and apply novel computer-extracted features characterizing the interaction and spatial co-localization of TILs and cancer nuclei (SpaTIL) in stratifying patients based on OS following nivolumab therapy.
Method
H&E tissue slides obtained from pre-treatment biopsies of 96 NSCLC patients treated with nivolumab were digitized and included for this study from 3 different institutions with the tumor region annotated by pathologists. Then 85 SpaTIL features related to TIL density, architecture and co-localization with tumor cells have been extracted to represent each patient. The most discriminative and uncorrelated features were selected by Elastic-Net regularized Cox-regression model to predict OS. The model was trained on D1 (n=25) and independently validated in D2 (n=32) and D3 (n=64). Multivariate analysis with clinico-pathologic factors was also performed.
Result
The top features consisted of the abundance of TILs around tumor cells and the distribution of the TILs. On the validation set, SpaTIL classifier yielded a HR=3.03 (95%CI=1.1 -8.35; p=0.042) on D2 and HR=4.12 (95%CI=1.87-9.09; p=0.02) on D3 by a log-rank test. On multivariate analysis with stage, smoking, histologic type, total lymphocyte count (See Table 1) SpaTIL was independently prognostic of OS (HR=7.88; 95%CI=1.66 – 37.216; p=0.009).
Table 1. Multivariate analysis for overall survival on the validation sets D2 and D3 Variables
HR(95% CI)
p value
Age (>65 vs <=65 yrs)
0.99(0.97-1.03)
0.67
Gender (Male vs Female)
1.05(0.75-2.79)
0.88
Smoking Status
(Former vs Never smoker)
3.19(0.92-11.061)
0.07
Histological Subtypes (Adeno vs Squamous)1
1.06(0.13-8.54)
0.95
EGFR status
1.32(0.49-3.52)
0.58
ALK status
0.63(0.36-1.10)
0.10
Total lymphocyte count
0.99(0.99-1.00)
0.33
SpaTIL Classifier
7.88(1.66-37.216)
0.009
CI = confidence interval; HR = Mantel-Haenszel Hazard ratio. Values in bold are statistically significant, p<=0.05.
Conclusion
Spatial interaction of TILs and cancer are independently prognostic of OS in nivolumab treated NSCLC. Further validation needs to be done to evaluate its utility.
Only Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.
-
+
P1.04 - Immuno-oncology (ID 164)
- Event: WCLC 2019
- Type: Poster Viewing in the Exhibit Hall
- Track: Immuno-oncology
- Presentations: 1
- Moderators:
- Coordinates: 9/08/2019, 09:45 - 18:00, Exhibit Hall
-
+
P1.04-25 - CT Based Vessel Tortuosity Features Are Prognostic of Overall Survival and Predictive of Immunotherapy Response in NSCLC Patients (ID 826)
09:45 - 18:00 | Author(s): Kaustav Bera
- Abstract
Background
Recently majority of patients with advanced non-small cell lung cancer (NSCLC) without targetable mutations are treated with immune checkpoint inhibitors (ICI). Since there are currently no validated biomarkers for predicting benefit of immunotherapy (IO), there is an unmet clinical need for development of such biomarkers. The tumor vasculature is a key component of the tumor micro-environment that can influence its behavior and therapeutic refractoriness. We aimed to evaluate the prognostic and predictive potential of quantitative vessel trotuosity (QVT), in the NSCLC patients treated with ICI drugs. Two hypotheses were established: first, the QVT on pre-treatment CT scans of NSCLC patients are associated with overall survival (OS). Second, the prognostic QVT features can lead to identify the patients who will benefit from IO.
Method
This study include 128 patients with advanced NSCLC. All patients underwent a baseline contrast CT imaging. Patients who did not receive IO drugs after 2 cycles due to a lack of response or progression as per RECIST were classified as non-responders. The dataset was splitted into a discovery (N=64) and validation sets (N=64). A set of 74 QVT features pertaining to tortuosity and curvature of tumor vasculature was extracted in CT scans. The initial set of QVT features were reduced to 8 features using least absolute shrinkage and selection operator (LASSO) in conjunction with OS data of the patients. Then, cox proportional hazard model was used to determine the contribution of each feature for categorizing survival groups. The weighted sum of selected 8 features gave a risk score (QRS) per patient. Patients in validation set were stratified based on QRS using the cutoff and feature weights learned in the discovery set. Prognostic features in conjunction with a linear discriminant machine learning model and OS were used to build a model to predict the response to IO. The prognostic features were also used for unsupervised clustering of the patients.
Result
The QRS risk score was able to stratify patients into two survival groups in validation set (Fig1. a-b) with p-value=0.022, Hazard ratio (HR)=0.47 and concordance index (CI)=0.61. The response prediction model yielded an AUC of 0.64±0.03 (Fig1.c). The agreement between patients with high OS and positive response to therapy was found to be 0.62 on unsupervised clustering method (Fig1. d).
Conclusion
The CT extracted QVT features was found to be prognostic of OS and also showed predictive value that could be used to identify patients who will benefit from IO.
-
+
P2.04 - Immuno-oncology (ID 167)
- Event: WCLC 2019
- Type: Poster Viewing in the Exhibit Hall
- Track: Immuno-oncology
- Presentations: 1
- Moderators:
- Coordinates: 9/09/2019, 10:15 - 18:15, Exhibit Hall
-
+
P2.04-16 - Novel CT Based Radiomic Features are Prognostic and Predictive of Benefit of Chemoimmunotherapy in Advanced Non-Squamous NSCLC (ID 2769)
10:15 - 18:15 | Presenting Author(s): Kaustav Bera
- Abstract
Background
Carboplatin, pemetrexed and pembrolizumab (C/P/P) is currently approved for patients with advanced non-squamous carcinoma of the lung (NS-NSCLC) based on superior survival outcomes noted in KEYNOTE-189. Since clinical benefit was observed across all PD-L1 expression categories, there are currently no robust predictive biomarkers that can identify subsets of patients likely to derive benefit from this regimen. We sought to evaluate whether radiomic features extracted from within and outside the nodule on pre-therapy CT scans could predict response to C/P/P.
Method
We retrospectively identified 52 patients with stage IV NS-NSCLC who received C/P/P. Of these, 6 were excluded because of non-evaluable thoracic lesions. Lung tumors were contoured on 3D SLICER software by an expert reader. Textural and shape radiomic features were extracted from intra/peritumoral regions using MATLAB® 2018b platform (Mathworks, Natick, MA). The primary endpoint of our study was RECIST response and secondary end point was overall survival (OS). A linear discriminant analysis classifier (LDA) was used to predict response across 100 iterations of threefold cross validation in the dataset. Performance of classifier on response was measured by area under receiver operating characteristic curve (AUC). To build the multivariate radiomic signature for OS, least absolute shrinkage and selection operator (LASSO) Cox regression model was used and a risk score was computed according to a linear combination of selected features. Patients were divided into high-risk or low-risk groups based on median risk score.
Result
The top five radiomic features (intra/peritumoral textural patterns) predictive of response to C/P/P were identified by mRMR feature selection method. LDA classifier using these features could discriminate responders from non-responders with an AUC of 0.77 ± 0.05.
The radiomic risk score was calculated using a linear combination of top six selected features from LASSO with corresponding coefficients. In a multivariate Cox proportional hazards model using a combination of clinicopathologic and radiomic features, the radiomics signature was found to be significantly associated with OS (averaged on 100 iteration of CV) (HR 10.42; 95% CI: 4.18-26; P = 4.92e-07). Kaplan-Meier survival analyses according to the radiomics signature risk-score showed significantly worse survival in the high risk category.
Conclusion
Textural features within and outside the nodule on pre-treatment CT images of patients with NS-NSCLC treated with C/P/P were predictive of responses and OS. Additional validation of these quantitative image-based biomarkers in independent cohorts is warranted.
Figure: Kaplan-Meier survival analyses of patients (N = 46) with NS-NSCLC treated with C/P/P using the radiomics signature risk-score.
-
+
P2.17 - Treatment of Early Stage/Localized Disease (ID 189)
- Event: WCLC 2019
- Type: Poster Viewing in the Exhibit Hall
- Track: Treatment of Early Stage/Localized Disease
- Presentations: 2
- Moderators:
- Coordinates: 9/09/2019, 10:15 - 18:15, Exhibit Hall
-
+
P2.17-34 - Integrated Clinico-Radiomic Nomogram for Predicting Disease-Free Survival (DFS) in Stage I and II Non-Small Cell Lung Cancer (ID 2886)
10:15 - 18:15 | Presenting Author(s): Kaustav Bera
- Abstract
Background
Early stage non-small cell lung cancer (ES-NSCLC) comprises about 45% of all NSCLC patients, with 5-year survival ranging between 30-49%. Surgical resection is the standard of care curative modality in these patients but about 30-55% of patients often recur following surgery within the first 3 years. There is currently no validated method to stratify patients based on their risk of recurrence following surgery in these patients. In this project, we develop and validate a nomogram using a combination of CT-derived radiomic textural features and clinco-pathologic factors, in order to predict DFS in ES-NSCLC.
Method
This study comprised 350 ES-NSCLC patients from two different institutions who underwent surgery (75 patients relapsed). Radiomic textural features were extracted from tumor region (Intratumoral - IT) as well as from the annular ring shaped peritumoral region (PT) with 3mm as a ring thickness and extending 9 mm outside the nodule. A total of 124 features from Gabor, Laws, Laplace, Haralick and Collage feature families were extracted from IT and each PT ring for all patients. The most stable, significant and uncorrelated features were selected from D1 (N=221) and used to build a Lasso-regularized multivariate Cox-regression model to generate a Radiomic Risk Score (RRS) derived from weighted Lasso coefficients. Further, RRS was integrated with clinic-pathologic variables (Lympho-vascular invasion LVI and AJCC stage) which were independently predictive on DFS in multivariate analysis to build a clinical-radiomics score (CRS). A nomogram was constructed to visually assess the CRS and RRS on DFS. Performances were evaluated using hazard ratios (HR), concordance index (C-Index) along with decision and calibration curves to show the differences between the individual and integrated risk scores.
Result
Top 14 radiomic features included 6 from IT and 8 from 0-9 mm PT distance. The constructed RRS could predict DFS (n=221, C-index=0.69, HR = 3.8; 95% CI- 2.7-5.6, p<0.05) on training (D1) and (n=129, C-index=0.69, HR – 2.5; 95% CI – 1.8-4.7; p<0.05) on a blinded validation cohort (D2), Addition of LVI and pN to build the CRS, increased C-index to 0.74, (p<0.05). Decision and calibration curve analysis shows improved performance of CRS over RRS or clinco-pathologic factors alone.
Conclusion
Addition of prognostic clinical factors (LVI, AJCC stage) improved the performance of the Radiomic Risk Score model in order to accurately predict DFS in ES-NSCLC patients undergoing surgery.
-
+
P2.17-35 - Integrating CT Radiomic & Quantitative Histomorphometric Whole Slide Image Features Predicts Disease Free Survival in ES-NSCLC (ID 2910)
10:15 - 18:15 | Author(s): Kaustav Bera
- Abstract
Background
Early-Stage non-small cell lung cancer (ES-NSCLC) accounts for approximately 40% of NSCLC cases, with 5-year survival rates varying between 31-49%. Radiomic textural features from pre-treatment CT scans and QH features from H&E stained WSIs have been shown to be independently prognostic of outcome. With diagnostic CT scans and surgical resection, the standard of care in ES-NSCLC, in this work we seek to take a multimodality approach using routine imaging to improve the predictive performance in determining DFS following resection.
Method
A retrospective chart review of Stage I and II (ES-NSCLC) pts undergoing surgical resection between 2005-14 with available CT and resected tissue yielded 70 pts. A total of 248 radiomic CT textural features from inside the tumor (Intratumoral –IT) and outside the tumor (Peritumoral – PT) and 242 QH features related to the nuclear shape, texture and spatial orientation and architecture from H&E WSI were extracted. We developed two risk models, Radiomic and QH using the most stable, discriminative and uncorrelated features from CT and WSI respectively determined by Lasso-regularized Cox regression to predict Disease free survival (DFS). Model performances were analyzed using Hazard Ratios (HR), Concordance Index (C-index) and Decision curve analysis. We built a nomogram to calculate the DFS based around the individual models as well as an integration of the QH and Radiomic models.
Result
Top 6 Radiomic features included 2 IT and 4 PT features from the Haralick and Collage families. The QH model comprised 6 nuclear shape and graph features. In predicting DFS, While the Radiomic model had a HR of 2.4 (p <0.01) with C-index – 0.67, the QH model had HR – 3.1 (p <0.01) with C-index – 0.74. Integration of the Radiomic and QH model yielded a C-index of 0.78 (p< 0.01). After addition of prognostic clinical factors (LVI, AJCC stage) to the model, the C-index was 0.80, almost doubling either modalities alone. The constructed nomogram visualized the apparent benefits of the three models while a decision curve clearly demonstrated the increased benefit of combined integrated model.
Conclusion
Integration of CT-derived radiomic and tissue-derived QH features was found to show improved performance in predicting RFS when compared to either radiomics or QH alone.
