Publications

External Validation of QVT Phenotypes: High Vascular Complexity Predicts Poor Survival in NSCLC

Radiomic phenotypes of tumor angiogenesis compared with PD-L1 in pre-treatment prediction of outcomes across immunotherapy regimens in NSCLC: An external validation study.,  ASCO Annual Meeting, June 2025

Multimodal AI Imaging Biomarker Integrating Radiomics, Pathomics, and Clinical Data Outperforms RECIST in Predicting Survival

Effect of fusion of radiomic, pathomic, and clinical biomarkers on multi-scale tumor biology and OS stratification in HNSCC receiving standard of care (SOC).,  ASCO Annual Meeting, June 2025

In patients with advanced HNSCC receiving standard-of-care immunotherapy, a biomarker combining radiomic, pathomic, and clinical features significantly outperformed RECIST and clinical biomarkers (e.g., PD-L1, P16) in stratifying overall survival. Developed using the Picture Health Px™ Platform, the multimodal biomarker was prognostic of OS (HR: 6.3, p=0.0001) with a median survival gap of over 4 years between high- and low-risk groups. Each data modality contributed independent prognostic value. This interpretable biomarker may better identify high-risk patients who would benefit from alternative therapies, enabling more personalized and effective HNSCC management.

QVT Phenotypes: High Vascular Complexity Predicts Poor ICI Outcomes in NSCLC

Novel radiologic phenotypes of chaotic tumor angiogenesis associated with poor ICI outcomes in NSCLC,  ESMO IO Congress, December 2024

Recent treatment successes suggest vascular complexity plays a crucial role in NSCLC outcomes.  Two distinct Quantitative Vessel Tortuosity (QVT) phenotypes (QVT-High, QVT-Low) were discovered from pre-treatment CT, quantifying complexity in the lesion-vascular network.  QVT-High, characterized by chaotic vascular structures, was significantly associated with worse overall survival (OS) and progression-free survival (PFS), independent of clinical variables and PD-L1 expression. These findings highlight a novel, non-invasive biomarker approach to guide patient selection for advanced anti-angiogenic  therapies and improve treatment monitoring. 

Multimodal AI Biomarker: A superior ICI response predictor over any single modality

Multimodal AI biomarker fusing radiology, pathology, and molecular information for immune checkpoint inhibitor (ICI) response prediction in lung adenocarcinoma (LUAD),  SITC Annual Meeting, November 2024

Traditional ICI biomarkers, like PD-L1, are insufficient to predict ICI response. Radiology and pathology contain complementary information to traditional molecular biomarkers, characterizing the tumor habitat at different scales. We have developed a multimodal AI biomarker, integrating radiology, pathology, and molecular data to predict ICI response. Trained and validated on late-stage LUAD patients across multiple datasets, the biomarker significantly outperformed unimodal models, achieving an AUC of 0.87 in an external test set and 0.90 in patients with all three data modalities available. Even with missing data, the model maintained robust predictive accuracy (AUC=0.81), demonstrating its potential to complement or surpass PD-L1 as a clinical decision-making tool.

Pathomic biomarker outperforms PD-L1 for ICI response prediction

External validation: Tumor-immune spatial interactions on NSCLC H&E slides predict immunotherapy response,  ESMO Annual Meeting, September 2024

Many patients, including those with high PDL1 expression, fail to respond to ICI treatment, suggesting the need for more precise biomarkers. To address this, a novel AI-based immunotherapy response (PIRe) score was developed. PIRe is based on tumor infiltrating lymphocyte density (denTIL) and their spatial interactions with tumor cells (spaTIL) and was trained and validated on  pre-treatment H&E stained slide images from late-stage NSCLC patients across three institutions. PIRe showed an AUC of 0.71 overall, 0.82 for ICI recipients, and 0.66 for chemo-ICI patients. When combined with PDL1 expression, PIRe improved predictive accuracy over PDL1 alone, with a combined AUC of 0.82: 0.96 for ICI recipients, and 0.85 for chemo-ICI patients. 

CheckpointPx: a better ICI response than PD-L1 alone

CheckpointPx, an interpretable radiology AI tool, predicts immune checkpoint blockade benefit independent of PDL1 status in non-small cell lung cancer (NSCLC): A multi-institutional validation study, AACR Annual Meeting, March 2024

CheckpointPx, an AI signature driven by interpretable imaging biomarkers, predicts response to ICI treatment regimens in NSCLC patients. This signature also significantly stratifies patients by Progression-free survival (PFS) (HR=1.67 [1.22-2.28], p=0.001. This separation remained significant within subsets of PD-L1 negative (HR=2.71 [1.35-5.44], p=0.005) and PDL1-positive patients (HR=2.05 [1.22-3.46], p=0.007), suggesting the potential of this novel signature to address critical gaps in the current NSCLC immunotherapy biomarker landscape.

Vessel tortuosity as a predictor of overall survival in NSCLC

A tumor vasculature–based imaging biomarker for predicting response and survival in patients with lung cancer treated with checkpoint inhibitors, Science Advances, November 2022

The predictive value of QVT and delta-QVT was assessed in lung cancer patients undergoing immunotherapy. Our findings demonstrate that QVT can predict patient response and overall survival, evaluated across three validation sets from separate institutions s (OS; HR = 2.49 [95% CI = 1.17 to 5.32], P = 0.002; HR = 2.12 [95% CI = 1.04 to 4.29], P = 0.014, and HR = 2.98 [95% CI = 1.12 to 7.93], and P = 0.04). Furthermore, delta-QVT emerges as a robust predictor of ICI response (RECIST, AUC = 0.92 and 0.85). Using QVT, we can select patients for immunotherapy trials more effectively, pinpointing those who are most likely to benefit from treatment. Notably, by assessing abnormal tumor vasculature, QVT can be leveraged as a biomarker for targeted therapies, including VEGF anti-angiogenic treatment. Detecting early changes in this structure can signal treatment effectiveness.

TIL density and distribution as predictor of benefit from IO

Spatial interplay patterns of cancer nuclei and tumor-infiltrating lymphocytes (TILs) predict clinical benefit for immune checkpoint inhibitors, Science Advances, June 2022

The predictive potential of Picture Health’s proprietary spaTIL and denTIL features was assessed in patients with advanced tumors receiving immune checkpoint inhibitor (ICI) treatment. These features, which quantify the interplay between tumor cells and tumor-infiltrating lymphocytes (TILs), significantly predicted survival outcomes in patients with NSCLC and gynecological cancer receiving ICI treatment, independent of clinical factors including PD-L1 expression. Picture Health’s digital pathology feature families, tethered to tumor biology, can provide valuable insights into tumor characteristics, enabling improved biomarkers and patient selection strategies.