Bhavani S. Singh

Macrophages are key contributors to vascularization, but the mechanisms behind their actions are not understood. Here, we show that diverse macrophage phenotypes have distinct effects on endothelial cell behavior, with resulting effects on vascularization of engineered tissues. In Transwell coculture, proinflammatory M1 macrophages caused endothelial cells to up-regulate genes associated with sprouting angiogenesis, whereas prohealing (M2a), proremodeling (M2c), and anti-inflammatory (M2f) macrophages promoted up-regulation of genes associated with pericyte cell differentiation. In 3D tissue-engineered human blood vessel networks in vitro, short-term exposure (1 day) to M1 macrophages increased vessel formation, while long-term exposure (3 days) caused regression. When human tissue-engineered blood vessel networks were implanted into athymic mice, macrophages expressing markers of both M1 and M2 phenotypes wrapped around and bridged adjacent vessels and formed vessel-like structures themselves. Last, depletion of host macrophages inhibited remodeling of engineered vessels, infiltration of host vessels, and anastomosis with host vessels.

Abstract Purpose: De-identification of cancer imaging data is vitally important for data sharing and the advancement of research, however it is a time consuming and complex process that limits access to new cancer data sets such as those shared through NCI's Imaging Data Commons (IDC), built on the Google Cloud Platform (GCP). Our research demonstrates how this process can be automated using GCP-native services. Methods: We configured the Medical Image De-Identification (MIDI) pipeline to automate de-identification of cancer imaging data. De-identification is performed using an alpha release of GCP’s Healthcare API which was configured to scrub all Protected Health Information (PHI) from both Digital Imaging and Communications in Medicine (DICOM) headers and burnt-in text in pixel data. A dataset containing 216 patients and 23,921 images was prepared to test the de-identification algorithm by placing synthetic PHI in both DICOM headers and pixel data. The synthetic data matched real data seen during curation at The Cancer Imaging Archive (TCIA) and included data difficult for an algorithm to detect. Accuracy of the MIDI pipeline was measured against TCIA’s standard tools and procedures for de-identification. Measures included correct detection of all PHI data and correct action taken (e.g., remove, encrypt, or otherwise obscure). Throughput was also measured. Results: Throughput was measured at 22.0 images per second over 10 runs. The MIDI pipeline’s accuracy for DICOM headers was 98.7%, accurately detecting dates, addresses, phone numbers, unique identifiers, names, and other common PHI. The most common PHI failed to remove were special cases that included uncommon names or names with symbols, dates in string data types that were mistaken for other IDs, patient IDs, and abbreviated institution names. Private Creator data elements were consistently failed to be retained. These errors were due to options not currently available, and algorithms not trained on specific PHI, such as abbreviated institution names. UIDs were correctly replaced. PHI burnt-in the pixel data was successfully detected and removed, with one false positive. Conclusion: We demonstrate the current capability and performance of automated cancer image de-identification. Our results show that while full automation is within grasp, a semi-automated pipeline is now feasible. A human expert in the loop can be used for final verification. This will lead to a much-needed acceleration of image de-identification, to handle the rapidly growing volume of data and provide rapid timely access in support of cancer research. Future work will focus on including pre- and post-processing tools to aid the human expert in the loop, such as identifying and flagging questionable images for manual review. These tools will also be used to catch the errors mentioned in results. Citation Format: Benjamin P. Kopchick, Laura K. Opsahl-Ong, Qinyan Pan, Michael W. Rutherford, Ulrike Wagner, Bhavani S. Singh, Scott Gustafson, Fred W. Prior, David A. Clunie, Juergen A. Klenk, Keyvan Farahani. Accelerating de-identification of images with cloud services to support data sharing in cancer research. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6579.

Abstract Purpose: The National Cancer Institute’s (NCI) Cancer Research Data Commons (CRDC) is a cloud-based data ecosystem that allows researchers to share and access clinical, genomic, proteomic, and imaging data. CRDC currently houses more than 10 petabytes (PB) of data (predominantly genomic). The volume of genomic data in CRDC has more than doubled since 2022 from 3.7 PB to 8.8 PB. To address the escalating data storage costs, CRDC must identify economic genomic data storage/compression strategies to achieve long-term sustainability. Methods: To evaluate the impact of genomic compression algorithms and cloud storage solutions, a data compression and storage pilot study was conducted based on two CRDC data sources:185GB from 1000 Genomes Project, and 151GB from the Integrated Canine Data Commons (ICDC). Four compression algorithms - PetaGene, CRAM, PigZ, and Genozip - were chosen based on their current use within the cancer research community. The study consisted of three parts: Compression-Only: four compression algorithms were evaluated for efficiency and cost-effectiveness Cloud Storage+Compression: compressed data were placed in various AWS S3 storage tiers including AWS Intelligent-Tiering (Assumption: no monitoring and automation costs for intelligent tiering) AWS HealthOmics: a range of possible storage costs were based on two scenarios- (1) data accessed monthly, and (2) data never accessed. (Assumptions: 4 gigabases per gigabyte, no egress costs when moving data off HealthOmics, each genome is downloaded in 500 parts generating 500 GET API calls) Results: Of the four algorithms tested, PetaGene performed best on both datasets. For 1000 Genomes data, PetaGene compressed data by 76% in ∼70 minutes for a one-time cost of $2.86. For ICDC data, PetaGene compressed data by 83% in ∼63 minutes for $2.57. It is broadly accepted that tiered storage yields cost savings. When compressed data was placed in tiered storage, yet more savings were realized. For both datasets, the most cost-effective strategy was S3 Intelligent-Tiering of PetaGene compressed data. The annual cost for PetaGene compressed data stored in intelligent tiering ranges from $2.10 - $6.90 (ICDC) and $3.70 - $12.14 (1000 Genomes). The annual cost to store data in AWS HealthOmics is $11.15 - $41.82 (ICDC) and $13.66 - $51.23 (1000 Genomes) for data accessed monthly. Conclusion: Significant cost savings can be achieved using effective genomic data compression tools paired with intelligent tiering storage solutions. There were two main limitations. Data license costs of compression algorithms were not studied, and while the frequency of data access should be considered for real world application, this was not part of the scope of this study. Both these factors will need to be considered as CRDC selects strategies to mitigate cost and inform overall infrastructure. Citation Format: Juergen Klenk, Dina Mikdadi, Bhavani Singh, Chelsea Owens, Eric Barner, Ross Campbell, Mary A. Sears, Ina Felau, Michael Warfe, Erika Kim, Tanja Davidsen. Analyzing economic storage solutions for cancer research data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1085.

Abstract Purpose: The National Cancer Institute’s (NCI) Cancer Research Data Commons (CRDC) supports the cancer research community by providing cloud-based, secure storage and analytic tools for multiple cancer data types (e.g., genomic, proteomic, imaging, and clinical trial data). CRDC is beginning to integrate artificial intelligence (AI) solutions, such as using AI to annotate medical images and making AI models and resources available for cancer researchers. As AI policies emerge, CRDC must track relevant guidelines to determine how AI can safely and compliantly unlock value for researchers. Methods: CRDC conducted a landscape analysis to inform CRDC of potential AI risks and policies across three branches of the Federal government and international agencies. CRDC identified and analyzed policies from organizations focused on AI risk mitigation measures, general AI guidance, and AI strategy as relates to federal health agencies. Results: The following policies have the most implications for CRDC. In the Executive Branch, the Executive Order on Safe, Secure, and Trustworthy AI mandates federal agencies to ensure AI safety and trustworthiness. The National Institute for Standards and Technology developed AI evaluation criteria, risk assessments, and tools to help assess AI models. CRDC can follow this mandate and use the tools to begin developing an AI strategy. In the Legislative Branch, the Senate AI Working Group and House AI Task Force are starting to discuss specific AI legislation. Although legislation hasn’t passed, CRDC should track legislative updates that could impact CRDC compliance. In the Judicial Branch, the Supreme Court overturned the Chevron Deference. Although not directly related to AI, the decision could affect how CRDC interprets and acts upon AI guidance. Internationally, European countries developed accountability frameworks, evaluation criteria, tools, and risk assessments to identify roles and responsibilities and help assess AI models. While NCI is not obligated to follow European AI policies, these artifacts can inform CRDC in developing an AI strategy. Conclusion: AI policies continue to emerge and evolve. As guidance is solidified, it is important that CRDC continues to monitor and analyze requirements to mitigate risks and ensure responsible use of AI in cancer research. This ongoing analysis will inform CRDC, enhance trust within their user community, increase timely access to cancer data analytical tools, and contribute to the advancement of cancer therapies. Citation Format: Juergen Klenk, Dina Mikdadi, Bhavani Singh, Chelsea Owens, Eric Barner, Ross Campbell, Mary Sears, Ina Felau, Erika Kim, Tanja Davidsen. Analyzing artificial intelligence (AI) policy for cancer research data programs [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 1084.