Vidya Sankar Viswanathan’s research while affiliated with Emory University and other places

Unresectable stage III NSCLC is now treated with chemoradiation (CRT) followed by immune checkpoint inhibitors (ICI). Pneumonitis, a common CRT complication, has heightened risk with ICI, potentially causing severe outcomes. Currently, there are no biomarkers to predict pneumonitis risk or differentiate between radiation-induced pneumonitis (RTP) and ICI-induced pneumonitis (IIP). This study analyzed 293 patients from two institutions, with 140 experiencing pneumonitis (RTP: 84, IIP: 56). Two models were developed: M1 predicted pneumonitis risk using seven radiomic features, achieving high accuracy across internal and external datasets (AUCs: 0.76 and 0.85). M2 differentiated RTP from IIP, with strong performance (AUCs: 0.86 and 0.81). Gene set enrichment analysis linked high pneumonitis risk to pathways such as ECM-receptor interaction and T-cell signaling, while high IIP risk correlated with MAPK and JAK-STAT pathways. Radiomic models show promise in early pneumonitis risk stratification and distinguishing pneumonitis types, potentially guiding personalized NSCLC treatment.

Artificial intelligence (AI) stands at the threshold of revolutionizing clinical oncology, with considerable potential to improve early cancer detection and risk assessment, and to enable more accurate personalized treatment recommendations. However, a notable imbalance exists in the distribution of the benefits of AI, which disproportionately favour those living in specific geographical locations and in specific populations. In this Perspective, we discuss the need to foster the development of equitable AI tools that are both accurate in and accessible to a diverse range of patient populations, including those in low-income to middle-income countries. We also discuss some of the challenges and potential solutions in attaining equitable AI, including addressing the historically limited representation of diverse populations in existing clinical datasets and the use of inadequate clinical validation methods. Additionally, we focus on extant sources of inequity including the type of model approach (such as deep learning, and feature engineering-based methods), the implications of dataset curation strategies, the need for rigorous validation across a variety of populations and settings, and the risk of introducing contextual bias that comes with developing tools predominantly in high-income countries.

8610 Background: CM227 (NCT02477826) is a large multi-center phase 3 trial that evaluated the benefit of immunotherapy (IO) over chemotherapy (Ch) as first line therapy in stage IV NSCLC. While PD-L1 and tumor mutational burden (TMB) initially emerged as promising biomarkers, CM227 showed benefit of IO over Ch regardless of PD-L1 or TMB status. There is an unmet clinical need for predictive biomarkers to identify patients (pts) who will respond to IO. In this study, we report initial blinded validation results of a CT-derived biomarker combining change in textural radiomics (Δ-Rad) and quantitative vessel tortuosity (Δ-QVT) between baseline and 6-week post-treatment for predicting response and survival outcome of IO and Ch alone in a subset of patients enrolled in CM227. Methods: This retrospective study included baseline (B) and 6-week post-treatment (TP1) CT scans from (a) a multi-center training set (D tr , N=110) of first line IO-treated mNSCLC pts and (b) a validation set consisting of a subset of mNSCLC from CM227 (D v , N=224), of which 178 pts were treated with IO (D v IO ) and 36 pts (D v Ch ) with Ch. Intra-tumoral, peri-tumoral texture radiomic (Khorrami et al., Cancer Immunol Res 2020) and QVT (Alilou et al., Sci Adv 2022) features were extracted from up to the two largest measurable lung lesions on each CT using an in-house MATLAB pipeline. Δ-Rad and Δ-QVT features were computed as the feature difference between B and TP1. A best objective IO response classifier, M Combo was trained on D tr using a combination of Δ-Rad and Δ-QVT features. Kaplan-Meier analyses with log rank p-values, hazard ratio (HR) and its confidence interval (CI) were computed to assess the predictive benefit of M Combo with overall survival (OS) and progression free survival (PFS) in D v , D v IO and D v Ch . We also report area under the receiver operating characteristic (AUC) of IO response predictions compared against best overall response in D v , D v IO and D v Ch . Results: M Combo predicted best overall response with an AUC of 0.67, 0.68, 0.74 in D v , D v IO , and D v Ch , respectively. M Combo was statistically significantly associated with OS and PFS in D v and D v IO but not in D v Ch (Table). Conclusions: Our preliminary findings reveal that combination of CT-based textural and vessel tortuosity features are predictive of IO response over chemotherapy in a subset of CM227 pts. Validation on the entire CM277 cohort is warranted. [Table: see text]

8600 Background: Immunotherapy (IO) has shown a durable response with minimal toxicity in the treatment of metastatic NSCLC. However, only 30-50% of these patients (pts) actually respond. Despite initial optimism surrounding PD-L1 expression as a potential biomarker, IO provides benefit across PD-L1 low (<25%) and high (≥25%) status. This highlights a critical need for more effective biomarkers to guide treatment decisions. Here, we present blinded validation results of a CT-based biomarker of change in quantitative vessel tortuosity (1) (Δ-QVT) and texture radiomics (2) (Δ-Rad) between baseline (B) and 6 weeks post-treatment (TP1) for predicting response, overall survival (OS) and progression free survival (PFS) in CP1108 (NCT01693562). Methods: CT scans at B and TP1 were retrospectively analyzed from two studies (a) a multi-center training set (S tr , N=110) of metastatic NSCLC pts treated with first line IO and (b) a blinded NSCLC cohort of N = 151 pts treated with Durvalumab in CP1108 study (S v ). In S v , N=75 pts were PD-L1 high (S v ⁺ ) whereas N=65 pts were PD-L1 low (S v ⁻ ). An in-house MATLAB based algorithm was used to extract intra-tumoral, peri-tumoral texture and QVT features from up to two largest measurable tumors in each CT. A classifier (M Combo ) was trained on S tr to predict best overall response defined as per the RECIST v1.1 criteria using combination of Δ-QVT and Δ-Rad features. Area under the receiver operating characteristic (AUC) was used to evaluate M Combo against RECIST response whereas univariable and multivariable Cox regression models with log-rank test, hazard ratio (HR) with confidence interval (CI) and concordance index (C-index) were used to study the association of M Combo with OS and PFS in S v , S v ⁺ , and S v ⁻ . Results: M Combo predicted objective response with an AUC of 0.78 in S v and was statistically significantly associated with PFS (Table) in S v , S v ⁺ , and S v ⁻ . M Combo was associated with OS in S v , S v ⁺ , but not in S v ⁻ . Multivariable analysis with PD-L1 status, histological subtype, ECOG status, liver metastasis, line of therapy and sex showed independent association of M Combo across OS and PFS. Conclusions: Our results indicate that combination of radiomics and vessel tortuosity biomarker from CT is independently prognostic of response and PFS across PD-L1 levels. Prospective validation of this biomarker is needed. 1. Alilou et al. Sci Adv 2022. 2. Khorrami et al. Cancer Immunol Res 2020. [Table: see text]

BACKGROUND Cardiac allograft rejection is the leading cause of early graft failure and is a major focus of postheart transplant patient care. While histological grading of endomyocardial biopsy samples remains the diagnostic standard for acute rejection, this standard has limited diagnostic accuracy. Discordance between biopsy rejection grade and patient clinical trajectory frequently leads to both overtreatment of indolent processes and delayed treatment of aggressive ones, spurring the need to investigate the adequacy of the current histological criteria for assessing clinically important rejection outcomes. METHODS N=2900 endomyocardial biopsy images were assigned a rejection grade label (high versus low grade) and a clinical trajectory label (evident versus silent rejection). Using an image analysis approach, n=370 quantitative morphology features describing the lymphocytes and stroma were extracted from each slide. Two models were constructed to compare the subset of features associated with rejection grades versus those associated with clinical trajectories. A proof-of-principle machine learning pipeline—the cardiac allograft rejection evaluator—was then developed to test the feasibility of identifying the clinical severity of a rejection event. RESULTS The histopathologic findings associated with conventional rejection grades differ substantially from those associated with clinically evident allograft injury. Quantitative assessment of a small set of well-defined morphological features can be leveraged to more accurately reflect the severity of rejection compared with that achieved by the International Society of Heart and Lung Transplantation grades. CONCLUSIONS Conventional endomyocardial samples contain morphological information that enables accurate identification of clinically evident rejection events, and this information is incompletely captured by the current, guideline-endorsed, rejection grading criteria.

Background Radiomics is a method within medical image analysis that involves the extraction of quantitative data from radiologic scans, often in conjunction with machine learning algorithms to phenotype disease appearance, prognosticate disease outcome, and predict treatment response. However, variance in CT scanner acquisition parameters, such as convolution kernels or pixel spacing, can impact radiomics texture feature values. Purpose The extent to which the parameters influence radiomics features continues to be an active area of investigation. In this study, we describe a novel approach, Acquisition Impact on Radiomics Estimation (AcquIRE), to rank the impact of CT acquisition parameters on radiomic texture features. Methods In this work, we used three chest CT imaging datasets (n = 749 patients) from nine sites comprising: i) lung granulomas and adenocarcinomas (D1) (10 and 52 patients, respectively); ii) minimal and frank invasive adenocarcinoma (D2) (74 and 145 patients); and iii) early-stage NSCLC patients (D3) (315 patients). Datasets D2 and D3 were collected from four sites each, and D1 from a single site. For each patient, 744 texture features and nine acquisition parameters were extracted and utilized to evaluate which parameters impact radiomic features the most. The AcquIRE method establishes a relative assessment between acquisition parameters and radiomic texture featuresa through the creation of a classification model, which is then utilized to assess the rank of the acquisition parameters. Results Across the use cases, CT software version and convolution kernel parameters were found to have the most variance. In D1, it was observed that the Haralick texture feature family was the least affected by variations in acquisition parameters, while the Gabor feature family was the most impacted. However, in datasets D2 and D3, the Gabor features were found to be the least affected. Our findings suggest that the impact on radiomic parameters is as much a function of the problem in question as it is acquisition parameters. Conclusions The software version and convolution kernel parameters impacted the radiomics feature the most.

Endometrial cancer (EC) disproportionately affects African American (AA) women in terms of progression and death. In our study, we sought to employ computerized image and bioinformatic analysis to tease out morphologic and molecular differences in EC between AA and European-American (EA) populations. We identified the differences in immune cell spatial patterns between AA and EA populations with markers of tumor biology, including histologic and molecular subtypes. The models performed best when they were trained and validated using data from the same population. Unsupervised clustering revealed a distinct association between immune cell features and known molecular subtypes of endometrial cancer that varied between AA and EA populations. Our genomic analysis revealed two distinct and novel gene sets with mutations associated with improved prognosis in AA and EA patients. Our study findings suggest the need for population-specific risk prediction models for women with endometrial cancer.