INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 264

Leveraging Vertex AI for Automated Ultrasound Image Analysis:

A Comprehensive Review

Hadi Khazaei*

1,4

, Behrooz Khajehee

, Danesh Khazaei

, Majd Oteibi

, Kaneez Abbas

1,4

, Bala Balaguru

Athreya Medtech

University of Milano-Bicocca

Portland State University

Validus Institute Inc

Corresponding Author

DOI: https://doi.org/10.51583/IJLTEMAS.2025.1408000033

Received: 01 Aug 2025; Accepted: 05 Aug 2025; Published: 29 August

Abstract: The integration of artificial intelligence (AI) into clinical ultrasound imaging presents profound opportunities for

advancing diagnostic accuracy, streamlining workflow efficiency, and ultimately improving patient outcomes. This comprehensive

review synthesizes contemporary literature and industry frameworks to evaluate current advances, practical use cases, observed

performance, and inherent limitations of Google Cloud's Vertex AI when applied to clinical ultrasound imaging studies.

The

platform's capabilities are not merely incremental; they represent a potential paradigm shift in how clinical ultrasound diagnostics

are approached and executed. This fundamental change can lead to reduced human variability in interpretation, optimized resource

allocation within healthcare systems, and faster patient diagnoses, though it also implicitly raises questions about associated

regulatory, ethical, and integration challenges that warrant careful consideration.

Keywords: Artificial intelligence (AI), Clinical ultrasound imaging, Vertex AI, Google Cloud, Machine learning in healthcare,

Medical imaging automation.

I. Introduction: The Evolving Landscape of AI in Clinical Ultrasound Imaging

Ultrasound imaging stands as a cornerstone of real-time, non-invasive diagnostics within clinical care, offering invaluable insights

for patient assessment across a multitude of medical disciplines.

Its widespread adoption is due to its safety, cost-effectiveness,

and ability to provide dynamic, live views of anatomical structures and physiological processes. Recent significant advancements

in artificial intelligence, particularly with sophisticated platforms like Google Cloud’s Vertex AI, are now enabling unprecedented

levels of automation in image annotation, disease detection, and overall workflow optimization within the medical imaging domain.

AI's integration into ultrasound medicine has revolutionized medical imaging, enhancing diagnostic accuracy and clinical

workflows.

By leveraging advanced algorithms such as convolutional neural networks (CNNs), AI has significantly improved

image acquisition, quality assessment, and objective disease diagnosis.

The strategic focus on enhancing "real-time, non-invasive diagnostics" through Vertex AI underscores a commitment to immediate

clinical utility and patient comfort. This emphasis suggests that Vertex AI's primary value proposition extends beyond post-hoc

analysis to directly support clinicians during or immediately after an examination. Such capabilities facilitate seamless integration

into live clinical workflows, offering immediate decision support. This is a significant advantage over AI tools that require extensive

offline processing, as it can directly contribute to reducing diagnostic delays and improving patient throughput in busy clinical

environments. By prioritizing real-time, non-invasive applications, Vertex AI is positioned as a tool that can democratize advanced

diagnostic capabilities, making sophisticated diagnostics more accessible and less burdensome for patients, particularly in scenarios

where rapid, non-invasive assessment is critical, such as emergency medicine or point-of-care ultrasound settings. AI can also be

instrumental in helping even novice users become more adept and confident in ultrasound utilization by providing continuous

feedback to improve image quality, help direct positioning, and intuitively guide users to clearer and more confident scans.

This review aims to comprehensively evaluate the current applications of Vertex AI in ultrasound imaging studies, assess its

observed effectiveness, and discuss the inherent challenges associated with its widespread adoption and integration into clinical

practice.

Vertex AI Platform: Capabilities and Clinical Relevance

Vertex AI is recognized as an end-to-end machine learning platform hosted on Google Cloud, distinguished by its comprehensive

suite of tools designed to manage the entire lifecycle of AI development and deployment.

This encompasses everything from initial

data handling to model interpretability, making it particularly relevant for complex clinical applications.

The platform's core capabilities, crucial for its utility in healthcare, include:

Data Management: Vertex AI facilitates robust data ingestion, meticulous curation, and essential de-identification processes, all

engineered to ensure strict compliance with stringent clinical privacy standards.

Vertex AI Vision, for instance, can quickly and

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 265

conveniently ingest real-time video and image streams at a global scale, and store and search petabytes of data with built-in AI

capabilities.

Model Development: The platform supports both automated machine learning (AutoML) and custom machine learning model

development, accommodating diverse data modalities including Vision and Tabular data, which are prevalent in medical imaging

and electronic health records.

Vertex AI Vision allows users to easily build computer vision applications using a drag-and-drop

interface, reducing development time from days to minutes.

For medical applications involving text generation, MedGemma, a

multimodal model, is pre-trained on a variety of de-identified medical data, including chest X-rays, dermatology images,

ophthalmology images, histopathology slides, medical text, medical question-answer pairs, and FHIR-based electronic health

record data.

Custom training jobs can be launched and tracked within the Vertex AI console, allowing users to specify container

settings and monitor progress.

Training and Deployment: Vertex AI provides robust tools for efficient model training, rigorous validation, and scalable

deployment. Its built-in scalability is designed to handle vast datasets and high computational demands inherent in large-scale

clinical imaging archives.

Explainability: The platform offers model explainability tools, which are increasingly recognized as crucial for achieving

regulatory approval and fostering clinical acceptance of AI-driven diagnostics.

The comprehensive nature of Vertex AI, spanning from data ingestion and de-identification to model development, training,

deployment, and explainability, suggests a strategic design aimed at addressing the entire complex lifecycle of AI development and

deployment within a clinical setting. In a clinical environment, the development and deployment of AI models often face significant

challenges due to fragmented tools, complex data governance, and the need for specialized expertise at each stage. Vertex AI's end-

to-end nature significantly reduces this complexity and fragmentation, streamlining the process. This makes AI development more

accessible and manageable for healthcare institutions that may not possess extensive in-house AI engineering teams, thereby

accelerating the translation of AI research into clinical utility. This integrated approach addresses a major bottleneck in clinical AI

adoption: the challenge of transitioning from isolated research prototypes to deployable, scalable, and maintainable solutions that

adhere to strict clinical standards. By offering a unified platform, Google Cloud aims to foster a more efficient, controlled, and

auditable environment for medical AI development, which is paramount for gaining trust and achieving regulatory approval. Vertex

AI has already demonstrated its utility across diverse healthcare contexts, including radiology, ophthalmology, and digital

pathology, and is rapidly gaining adoption in real-time imaging modalities such as ultrasound.

Applications of Vertex AI in Clinical Ultrasound

Vertex AI’s capabilities translate into several practical applications within clinical ultrasound imaging, offering solutions for various

diagnostic and workflow challenges.

Model Development and Auto ML Workflows

A significant advantage of Vertex AI lies in the ease with which machine learning solutions for image classification and anomaly

detection can be deployed using its AutoML tools. This often negates the requirement for deep coding expertise, democratizing AI

development within the healthcare sector.

Researchers successfully utilized AutoML on Vertex AI to classify data from the Breast

Imaging and Omics for Non-Invasive Integrated Classification (BIONIC) project. They developed a non-invasive classification

system for breast lesions, achieving precision and recall metrics comparable to those obtained from custom deep learning solutions.

The success of AutoML in ultrasound imaging diagnosis demonstrated that it can match the performance of tailored deep learning

approaches without requiring extensive coding, signifying a substantial democratization of AI development. This implies that high-

quality, clinically relevant AI models can be developed by domain experts, such as clinicians or medical researchers, who possess

profound medical knowledge but may not be specialized AI engineers. This "low-code" approach dramatically lowers the technical

barrier to entry for AI development in healthcare, empowering medical professionals to directly build, test, and iterate on AI models

tailored to their specific clinical needs. This can potentially accelerate the pace of innovation and ensure that models are more

clinically relevant and contextually appropriate, while also reducing reliance on external AI development teams, making the process

more agile and cost-effective for healthcare organizations. This trend suggests a future where clinical AI development is

increasingly driven by those with direct clinical insight, fostering a more integrated and efficient approach between clinical and

technical teams, ultimately benefiting patient care through faster AI adoption.

The typical workflow for developing models on Vertex AI involves several streamlined steps:

1. Uploading and labeling ultrasound images, which can be in formats such as DICOM or PNG.

2. Systematically splitting datasets into training, validation, and testing sets to ensure robust model evaluation.

3. Running multiple training iterations, with options to optimize node hours and apply early stopping to prevent overfitting.

4. Automatically evaluating key performance metrics, including AUPRC (area under the precision-recall curve), precision,

recall, and confusion matrices.

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 266

Integration with Medical Imaging Pipelines

A key feature of Vertex AI in clinical environments is its integration with the Google Cloud Medical Imaging Suite. This integration

enables seamless ingestion, annotation, privacy compliance, and deployment of complex image pipelines.

The Medical Imaging

Suite supports the convergence of diverse storage formats to the DICOM standard and offers seamless integration with on-premises

storage solutions.

It also provides an AI-assisted annotation environment, which is critical for efficiently processing and managing

the vast and continuously growing clinical ultrasound datasets, ensuring data integrity and accessibility throughout the AI lifecycle.

Illustrative Clinical Use Cases

Two internal AutoML experiments were conducted by our group using Google Cloud’s Vertex AI platform. These experiments

demonstrate how data quantity, annotation quality, and imaging modality influence model performance. By comparing the results

of a breast ultrasound project with those of an orbital ultrasound project, we highlight factors that contribute to success in real‑world

deployments.

Case 1 – BIONIC: AI‑assisted classification of cystic versus solid breast lesions

Figure 1: Phantom Breast Tissue

The Breast Imaging and Omics for Non‑Invasive Integrated Classification (BIONIC) project leverages portable Doppler ultrasound

imaging, phantom‑generated synthetic images, and curated annotations to develop a non‑invasive classification system for breast

lesions. We assembled a dataset of 93 breast ultrasound images annotated by experts as either cystic or solid. This dataset was

randomly split into 74 training, 10 validation, and 9 test images (approximately 80/11/9 %). A single‑label image classifier was

trained using Vertex AI AutoML Vision.

Figure 2: Left breast, solid lesion ultrasound image - using the synthetic tissue model.

Figure 3: Left breast, cystic lesion ultrasound image- using the synthetic tissue model.

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 267

Evaluation on the held‑out test set yielded an average precision (PR AUC) of 0.958 and precision and recall of 77.8 % at a

confidence threshold of 0.5. The confusion matrix showed that all cystic lesions were correctly classified, while 60 % of solid

lesions were correctly identified and 40 % were misclassified as cystic. These results demonstrate that larger datasets with

high‑quality annotations and synthetic phantom images can achieve robust performance using low‑code AutoML tools, providing

a strong foundation for clinical translation.

Figure 4: Vertex AI evaluation metrics for the BIONIC project.

Figure 5: Confusion matrix for the BIONIC Auto ML classifier

Case 2 – OID vs NIO: classification of orbital inflammatory disease

The OID vs NIO focused on differentiating Orbital Inflammatory Disease (OID) from Non‑Inflammatory Orbitopathy (NIO) using

portable point‑of‑care B‑scans.

Figure 6: Butterfly iQ+ handheld ultrasound device used in the OID vs NIO & BIONIC project.

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 268

Figure 7: Example of a normal orbital B‑scan image (NIO) from the OID vs NIO dataset.

Figure 8: Example of an annotated abnormal orbital B‑scan image (OID) from the OID vs NIO dataset.

A total of 100 orbital ultrasound images were collected, with 80 used for training, 10 for validation, and 10 for testing. The confusion

matrix revealed that 86 % of OID images were correctly classified.

Figure 9: Confusion matrix for the OID vs NIO AutoML model.

Correctly identifying NIO images was challenging, underscoring the limitations of small, imbalanced datasets and outdated

annotations. Nevertheless, this confirms that Vertex AI enables rapid prototyping and provides a baseline for future improvements.

We are currently expanding the OID vs NIO dataset with additional normal images and updated labels to enhance performance.

These two case studies underscore the importance of dataset size, annotation quality, and imaging modality in determining AutoML

performance. The BIONIC experiment demonstrates that well‑curated datasets with sufficient representation of each class can yield

high performance, whereas the OID vs NIO highlights the challenges of small, imbalanced datasets. Collectively, they illustrate

how Vertex AI empowers rapid AI development while emphasising that data quality remains the primary determinant of clinical

success.

Performance Analysis of Vertex AI Models in Ultrasound Diagnostics

The performance of Vertex AI models across various clinical imaging tasks demonstrates their efficacy and potential for widespread

adoption. The table below provides such a summary and highlights the clinical relevance of these AI‑assisted ultrasound

applications.

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 269

Figure 10: Precision–recall by curve & Precision–recall by threshold for the OID vs NIO model

Figure 11: Precision–recall by curve & Precision–recall by threshold for the BIONIC model

Table 1. Comparison of Vertex AI AutoML experiments (BIONIC vs OIDvsNIO)

Experime

Modality

Labels

Images

(train/val/test;

total)

AUC

Precisio

Reca

Confusion

matrix @0.5

Notes

BIONIC

Breast US

Cystic

vs Solid

74 / 10 / 9; 93

0.95

77.8%

77.8

Cystic100%,

Solid 60%

Curated + phantom

images

OIDvsNIO

Orbital B-

scan

OID vs

NIO

80 /10 / 10; 100

0.62

66.7%

66.7

OID 86%,

NIO 0%

Small, imbalanced,

real-patient set

These two experiments illustrate both the promise and the current limitations of low‑code AI solutions for ultrasound. The BIONIC

project, with its larger and better‑annotated dataset, achieved high accuracy and balanced performance across classes. In contrast,

the OIDvsNIO yielded moderate results due to its small size and severe class imbalance. Collectively, these internal studies

emphasise that data quantity and quality are the primary determinants of AutoML performance and that carefully curated,

diverse datasets are essential for clinically meaningful results.

Advantages of Adopting Vertex AI in Clinical Ultrasound Workflows

The adoption of Google Cloud’s Vertex AI in clinical ultrasound workflows offers several compelling advantages that address

critical needs in modern healthcare:

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 270

Accessibility: Vertex AI’s low-code AutoML capabilities significantly lower the barrier to entry for AI development. This

empowers healthcare professionals, including clinicians and medical researchers, to develop robust imaging models without

requiring extensive deep coding expertise, fostering greater innovation directly from domain experts.

Scalability: The platform is engineered for exceptional efficiency, allowing for the rapid and effective training and deployment of

AI models across vast and ever-growing imaging archives. This inherent scalability is crucial for managing large-scale clinical data

and supporting the demands of high-throughput diagnostic environments.

Compliance: Vertex AI integrates essential privacy and de-identification pipelines directly into its framework, ensuring that clinical

data handling adheres to stringent regulatory standards and patient privacy requirements, thereby mitigating significant legal and

ethical risks.

Google Cloud explicitly states that user interactions and content within Gemini (a generative AI model on Vertex AI)

stay within the organization and are not used for training models outside the user's domain without permission.

Google Cloud also

offers Business Associate Agreements (BAA) for handling Protected Health Information (PHI) under HIPAA, and holds

certifications like ISO 27001, ISO 27017, ISO 27018, ISO 27701, SOC 1, SOC 2, and SOC 3, demonstrating a commitment to

security and compliance.

Versatility: The platform demonstrates broad applicability, supporting a wide range of downstream applications beyond basic

classification. These include advanced tasks such as image segmentation, anomaly detection, and comprehensive workflow

orchestration within complex clinical environments, enhancing its utility across diverse medical specialties.

Economic Benefits and Efficiency Gains: AI in healthcare, including in diagnostic imaging like ultrasound, is projected to cut

healthcare spending by 5-10%, potentially saving $200 billion to $360 billion annually in the U.S. based on 2019 dollars.

These

savings stem from AI's ability to aid early diagnosis, facilitate correct treatment plans, and streamline administrative tasks.

Specifically, AI-driven ultrasound can reduce delays and errors in paperwork, leading to more complete and accurate reports and

fewer rejected insurance claims.

It can also automate repetitive image preparation tasks, reducing radiologist burnout and allowing

them to handle more cases.

Studies have shown reading time reductions of up to 52.57% and contouring time improvements

between 30-50% with AI integration.

AI also optimizes front-office operations by handling simple calls and connecting with

electronic health records (EHRs) and billing systems, reducing manual data errors and speeding up financial processes.

The synergistic combination of accessibility through low-code AutoML and scalability for efficient training and deployment across

vast imaging archives suggests that Vertex AI is strategically designed to overcome two of the most significant practical barriers to

AI adoption in healthcare: the scarcity of specialized AI talent and the immense, ever-increasing volume of medical data. Many

healthcare systems globally struggle with both a critical shortage of specialized AI/ML engineers and the overwhelming,

continuously growing volume of medical image data. Vertex AI's design directly tackles these issues by allowing non-experts, such

as clinicians and medical researchers, to build and iterate on models, and then providing the robust, cloud-based infrastructure

necessary to handle and process massive datasets efficiently. This implies a significant reduction in the operational overhead and

specialized expertise traditionally required to implement AI at scale within a clinical environment. This strategic combination of

features positions Vertex AI as a practical, sustainable, and democratizing solution for integrating AI into routine clinical practice,

suggesting a future where AI is not just a research curiosity but an integral, manageable, and widely accessible part of healthcare

operations. This could lead to widespread AI adoption even in resource-constrained settings, fostering a paradigm shift in how

clinical data is leveraged for diagnostics, patient management, and overall healthcare delivery.

Challenges and Limitations for Widespread Clinical Implementation

Despite its significant advantages, the widespread adoption and full clinical integration of Vertex AI in ultrasound imaging face

several critical challenges and inherent limitations:

External Validation Deficiencies: A common and significant limitation noted in current studies is the frequent absence of robust

external validation. This significantly restricts the generalizability of models across diverse patient populations, different clinical

settings, or various ultrasound devices, posing a substantial challenge for broader applicability in heterogeneous real-world

environments.

The recurring themes of "missing external validation" and "dataset constraints," particularly the lack of diversity

and exclusive reliance on internal datasets, point to a fundamental and pervasive challenge in medical AI: the "domain shift"

problem and the inherent difficulty of ensuring model generalizability across varied clinical environments. This means that an AI

model trained on data from one specific institution, patient demographic, or type of ultrasound machine may not perform reliably

or accurately when applied to data from different hospitals, diverse patient populations, or varying equipment. This is a major

roadblock for widespread clinical deployment. A model that demonstrates excellent performance in a controlled research lab or

within a single hospital's dataset is not clinically useful if its accuracy degrades significantly in diverse real-world settings. This

implies that while Vertex AI provides powerful tools for model development, the quality and representativeness of the input data

remain the determinants of a model's real-world utility, trustworthiness, and safety, necessitating significant collaborative effort in

data collection, sharing, and standardization across multiple institutions. The primary bottleneck for clinical AI adoption shifts from

purely technical model development challenges to broader issues of data infrastructure, data governance, and the establishment of

large, diverse, and externally validated medical imaging datasets. This implies that for AI to truly revolutionize clinical practice,

there needs to be a systemic shift towards multi-institutional collaborations or the adoption of privacy-preserving techniques like

federated learning to build robust, generalizable models.

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 271

Explainability Gaps: While explainability tools exist within the broader Vertex AI platform, some AutoML workflows have been

observed to lack natively built-in explainable AI capabilities.

This presents a significant challenge for achieving regulatory

acceptance and fostering clinical trust, as clinicians require transparency to confidently integrate AI-driven decisions into patient

care.

Without native explainability, complex manual post-hoc analyses are often necessitated to interpret model decisions in

critical clinical scenarios, adding an additional layer of complexity and time.

Dataset Constraints and Bias: The performance and reliability of Vertex AI models are highly dependent on the size, quality, and

diversity of the training data utilized.

Exclusive reliance on internal, potentially homogenous, datasets may severely limit the

broader applicability and robustness of the models when deployed in varied real-world clinical contexts, as they may not generalize

well to unseen data characteristics.

Bias, defined as systematic errors leading to a distance between prediction and truth, can be

introduced during dataset collection and preparation.

This can stem from demographic imbalances (e.g., if a dataset predominantly

consists of images from a particular racial or ethnic group, gender, age, or socio-economic status), leading to reduced accuracy for

underrepresented groups.

Variations in image quality and source (e.g., different institutions, equipment, protocols) can also

introduce bias, causing models to overfit to specific training data characteristics and reducing generalizability.

Furthermore, the

process of labeling medical images by human experts can lead to inconsistencies and annotation bias.

Statistical Robustness Assessment: Automated pipelines within Vertex AI may not natively support the generation of confidence

intervals or advanced metric variability assessment.

This lack of inherent statistical rigor can make it challenging for clinicians to

fully understand the uncertainty associated with an AI's prediction, thereby impacting diagnostic confidence and potentially

hindering regulatory approval processes. The identified lack of native support for "confidence intervals or advanced metric

variability assessment" in automated pipelines, when viewed in conjunction with the existing explainability issues, suggests a

critical gap in the platform's ability to provide the level of statistical rigor and interpretability typically required for high-stakes

clinical decision-making and regulatory approval. Clinicians and regulatory bodies require more than just a single point estimate of

an AI model's performance; they need to understand the model's uncertainty and the variability of its predictions across different

scenarios. In clinical practice, diagnostic certainty and understanding the range of possible outcomes are paramount for safe and

effective patient management. If an automated AI pipeline provides only a single, deterministic prediction or performance metric

without quantifying its uncertainty, it becomes significantly harder for clinicians to trust the model's output, especially in ambiguous

or critical cases. Similarly, regulators demand robust statistical evidence of reliability. This limitation forces healthcare

professionals to perform manual, post-hoc statistical analyses, which negates some of the efficiency benefits promised by

automation and adds complexity to the validation process. This points to a crucial need for AI platforms to evolve beyond simply

providing high-accuracy metrics to offering comprehensive statistical and uncertainty quantification. For AI to be truly integrated

into clinical decision support, it must not only provide an answer but also a transparent measure of its confidence in that answer,

allowing clinicians to appropriately weigh the risks and benefits. This is a critical area for future development to bridge the gap

between AI research and practical clinical utility, moving towards AI that explicitly communicates its limitations and uncertainties

to the end-user.

Ethical Considerations: Beyond technical limitations, the integration of AI in medical imaging raises several ethical concerns.

These include the privacy of data subjects, ensuring data quality and model efficacy, promoting fairness toward marginalized

populations, and maintaining transparency of clinical performance.

Proactively identifying and addressing AI bias is essential to

prevent negative consequences in clinical settings and ensure health justice.

Adoption Challenges and Practical Considerations

While technical limitations and ethical considerations are major hurdles, the real‑world adoption of AI‑assisted ultrasound tools

also hinges on practical barriers faced by clinicians and healthcare institutions. A recent global survey (COMPASS‑AI) of 1,154

healthcare professionals revealed that training and education are the most frequently cited barriers to using AI in point‑of‑care

ultrasound; 27.1 % of respondents identified the need for better training and guidance to effectively use AI tools. The same survey

highlighted a broader need for clinical validation and evidence, with 17.5 % of respondents noting a lack of robust data

demonstrating AI performance across different specialties and regions. These findings underscore that enthusiasm for AI does not

equate to readiness to adopt it—clinicians require accessible training programs and credible evidence of clinical benefit before

integrating AI into their workflow.

Practical obstacles also center on workflow integration and infrastructure. Clinicians report reluctance to use AI tools when there

is no standardized training curriculum or certification, and when models do not integrate smoothly with electronic medical records

and existing practice patterns. For AI to be useful at the point of care, it must fit seamlessly into existing care pathways without

adding undue burden on providers or requiring extensive redesign of current processes. This may necessitate collaboration between

developers, vendors, and clinical users to ensure that AI outputs are delivered in a way that is intuitive and relevant.

Cost is another significant barrier to adoption. Although AI ultrasound may offer long‑term savings through improved efficiency,

the initial investment can be substantial. Hospitals and clinics must purchase or lease software licenses, hardware (such as

ultrasound probes and computing infrastructure), integration services, staff training, and ongoing maintenance. Smaller facilities

may find these upfront costs prohibitive, limiting early adoption. Cost‑benefit analyses that quantify both initial expenses and

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025

www.ijltemas.in Page 272

long‑term return on investment can help decision makers justify funding. Shared or subscription‑based models may also lower

financial barriers for community hospitals and rural clinics.

Finally, user acceptance and trust play a crucial role in successful deployment. Healthcare professionals are understandably cautious

about over‑reliance on AI and may worry about job displacement or being forced to use opaque algorithms. Establishing clear

guidelines for AI use, emphasizing that AI serves as a decision support tool rather than a replacement for clinician expertise, and

promoting ongoing evaluation of AI systems can help build confidence in these technologies.

These adoption challenges reinforce that the success of AI‑assisted ultrasound depends not only on technical performance but also

on addressing human and organizational factors. Comprehensive training, transparent validation, thoughtful workflow integration,

flexible funding models, and stakeholder engagement will be critical to realizing the full potential of Vertex AI and similar

platforms in clinical practice.

Future Directions and Opportunities for Vertex AI in Ultrasound

The strategic roadmap for Vertex AI’s evolution in clinical ultrasound directly addresses the identified limitations and points

towards new frontiers in diagnostic and interventional applications:

Enhanced Explainable AI Integration: A crucial future direction involves the enhanced integration of explainable AI (XAI)

capabilities directly within the Vertex AI platform.

This will foster greater clinical trust, facilitate regulatory acceptance by

providing transparent insights into AI-driven decisions, and empower clinicians with a deeper understanding of the model's

reasoning.

Promotion of Open-Access Data and External Validation: There is a pressing need for more open-access datasets and rigorous

external validations across diverse populations and devices.

This will help standardize performance reporting across different

studies and institutions, significantly improving model generalizability and reliability, which are essential for widespread clinical

deployment.

Collaborative multi-center efforts are emphasized to generate diverse, standardized datasets.

Expansion into Advanced Modalities and Real-Time Guidance: Future developments are expected to include the expansion of

Vertex AI's capabilities into 3D ultrasound modalities and real-time guidance applications.

This is anticipated with ongoing

improvements in Google's model garden offerings and foundational models, opening new avenues for interventional and procedural

support, moving beyond purely diagnostic applications to more active clinical assistance. Vertex AI Vision is designed to ingest

real-time video and image streams, enabling the creation of computer vision applications for real-time insights.

AI-guided

ultrasound is already transforming healthcare by democratizing access to diagnostic-quality imaging, empowering clinicians with

limited ultrasound experience to capture precise images by providing real-time feedback on probe positioning and adjustments.

This technology reduces the skill barrier in ultrasound and supports the growing adoption of point-of-care ultrasound (POCUS).

Broader Google Cloud AI Roadmap: Google Cloud's future AI trends for healthcare include multimodal AI to unleash the power

of context, AI agents (like Google Agentspace) for scaling experimentation and deployment, optimization of the AI stack for

maximizing value, and "silo busting" to democratize AI access across departments.

Google's approach to AI development is

grounded in principles of bold innovation and responsible development, emphasizing human oversight, safety research, mitigating

unintended outcomes and bias, and promoting privacy and security.

These future directions explicitly address the major limitations identified previously, indicating a clear and responsive path for

Vertex AI to mature into a more clinically robust, trustworthy, and widely accepted platform. Prioritizing enhanced explainability

and advocating for broader data validation indicates a strategic commitment to building a platform that not only performs well but

also meets the stringent requirements of clinical medicine and regulatory bodies for transparency and reliability. The ambition to

expand into 3D ultrasound and real-time guidance signifies an intent to move beyond diagnostic AI to more complex, interventional,

and impactful applications. This forward-looking perspective reinforces the idea that Vertex AI is a continuously evolving platform,

adapting to the dynamic needs of medical imaging. It implies that as these future directions are realized, the platform's utility,

trustworthiness, and scope of application in clinical ultrasound will significantly increase, potentially accelerating the pace of AI

integration into routine clinical practice and broadening its impact on patient care. This also positions Vertex AI as a responsive

technology, capable of adapting to emerging clinical demands and technological advancements in the field of medical imaging.

Conclusion

Google Cloud’s Vertex AI facilitates significant advancements in the development and deployment of AI models specifically

tailored for clinical ultrasound imaging.

Its effectiveness is largely attributable to its user-friendly AutoML capabilities, robust

privacy compliance features, inherent scalability, and seamless integration capacity within existing clinical infrastructures.

These

attributes position it as a powerful tool for enhancing diagnostic accuracy and streamlining clinical workflows, while also offering

substantial economic benefits through efficiency gains and cost reductions.

However, while promising, addressing critical challenges such as enhancing explainability, ensuring data diversity, mitigating bias,

and conducting rigorous external validation remains paramount for achieving widespread clinical adoption and securing necessary

regulatory acceptance.

The conclusion's deliberate emphasis on balancing Vertex AI's demonstrated effectiveness with the crucial

need to address its identified limitations highlights the nuanced and complex reality of AI adoption in high-stakes fields like

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS) 
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025 
www.ijltemas.in                                                                                                                                                 Page  273 
 
medicine.  This  balanced  perspective  is  vital  for  an  expert  report,  as  it  acknowledges  the  substantial  progress  made  while 
simultaneously setting realistic expectations for current capabilities and outlining future requirements. It implies that while the core 
AI technology within Vertex AI is powerful, its successful and safe integration into routine clinical practice is contingent upon 
overcoming broader systemic challenges related to trust, regulatory frameworks, data governance, and the practical acceptance by 
clinicians. This nuanced conclusion reflects the broader  landscape of AI in healthcare, where technological  prowess  alone is 
insufficient for widespread impact. It underscores that successful clinical AI requires a holistic approach that considers not only 
technical performance but also ethical considerations, robust regulatory pathways, scalable data infrastructure, and genuine clinician 
acceptance.  This  reinforces  the  idea  that  the  future  of  AI  in  medicine  is  a  collaborative,  multi-stakeholder  effort  involving 
technologists, clinicians, policymakers, and patients, all working towards a common goal of improved patient care. 
With ongoing advancements and a focused effort on addressing these key areas, Vertex AI is well-positioned to further enhance 
AI’s pivotal role in improving ultrasound-based diagnostics and ultimately elevating patient care standards.
1
 
Acknowledgement 
We would like to extend our thanks to Professor Faryar Etesami (Portland State University), Adam Tamimi (Validus Institute), and 
Gabriel Tamimi (Validus Institute) for their contributions to this article. 
Bibliography 
1.  Oteibi, M., Tamimi, A., Abbas, K., Tamimi, G., Khazaei, D., & Khazaei, H. (2025). Breast tumor ultrasound: Clinical 
applications, diagnostic features, and integration with AI. International Journal of Research and Scientific Innovation, 
12(1). 
https://doi.org/10.51584/IJRIAS.2025.01001 
2.  Yi, Sung & Khazaei, Hadi & Khazaei, Danesh  & Junejo, Nashrah &  Etesami, Faryar. (2024). Unveiling the Future: 
Convergence  of  Engineering,  Medicine,  and  Technology  in  Biomedical  and  Biotechnology.  Medical  and  Clinical 
Research:  Open  Access.  5.  10.52106/2766-3213.1049.  and  Biotechnology 
https://www.researchgate.net/publication/388065365_Unveiling_the_Future_Convergence_of_Engineering_Medicine_a
nd_Technology_in_Biomedical_and_Biotechnology 
3.  Fundamentals of Orbital Inflammatory Disorders by Hadi Khazaei, https://link.springer.com/book/10.1007/978-3-031-
85768-3 
4.  Progress in the Application of Artificial Intelligence in Ultrasound-Assisted Medical Diagnosis - PMC - PubMed Central, 
accessed July 30, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC11939760/ 
5.  Progress in the Application of Artificial Intelligence in Ultrasound-Assisted Medical Diagnosis - MDPI, accessed July 30, 
2025, https://www.mdpi.com/2306-5354/12/3/288 
6.  AI  for  GPs:  How  AI  ultrasound  tools  are  changing  the  game  for  general  practitioners,  accessed  July  30,  2025, 
https://www.gehealthcare.co.uk/insights/article/ai-for-gps-how-ai-ultrasound-tools-are-changing-the-game-for-general-
practitioners 
7.  Vertex AI Vision | Google Cloud, accessed July 30, 2025, https://cloud.google.com/vertex-ai-vision 
8.  MedGemma  –  Vertex  AI  -  Google  Cloud  Console,  accessed  July  30,  2025,  https://console.cloud.google.com/vertex-
ai/publishers/google/model-garden/medgemma 
9.  Training  custom  models  on  Vertex  AI  -  YouTube,  accessed  July  30,  2025, 
https://www.youtube.com/watch?v=VRQXIiNLdAk 
10.  Medical Imaging Suite | Google Cloud, accessed July 30, 2025, https://cloud.google.com/medical-imaging 
11.  pmc.ncbi.nlm.nih.gov,  accessed  July  30,  2025, 
https://pmc.ncbi.nlm.nih.gov/articles/PMC11705385/#:~:text=AI%20enables%20automated%20echocardiographic%20i
mage,%2C%20increasing%20accessibility%2C%20and%20reproducibility. 
12.  Revolutionizing Cardiology: The Role of Artificial Intelligence in Echocardiography  - MDPI, accessed July 30, 2025, 
https://www.mdpi.com/2077-0383/14/2/625 
13.  AI  in  OB/GYN  Ultrasound,  accessed  July  30,  2025,  https://www.womens-health.net/education/blog/ai-in-obgyn-
ultrasound 
14.  Prospective Applications of Artificial Intelligence in fetal medicine | IJGM - Dove Medical Press, accessed July 30, 2025, 
https://www.dovepress.com/prospective-applications-of-artificial-intelligence-in-fetal-medicine--peer-reviewed-fulltext-
article-IJGM 
15.  Applications  of  artificial  intelligence  in  abdominal  imaging,  accessed  July  30,  2025, 
https://utsouthwestern.elsevierpure.com/en/publications/applications-of-artificial-intelligence-in-abdominal-imaging 
16.  Generative  AI  in  Google  Workspace  Privacy  Hub,  accessed  July  30,  2025, 
https://support.google.com/a/answer/15706919?hl=en 
17.  Compliance  and  security  controls  |  AI  Applications  -  Google  Cloud,  accessed  July  30,  2025, 
https://cloud.google.com/generative-ai-app-builder/docs/compliance-security-controls 
18.  Exploring the Economic Benefits of AI in Healthcare: Cost Reductions and Efficiency Gains, accessed July 30, 2025, 
https://www.simbo.ai/blog/exploring-the-economic-benefits-of-ai-in-healthcare-cost-reductions-and-efficiency-gains-
3576781/ 

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS) 
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue VIII, August 2025 
www.ijltemas.in                                                                                                                                                 Page  274 
 
19.  The  Economic  Impact  of  Artificial  Intelligence  in  Medical  Imaging  -  IDEAS/RePEc,  accessed  July  30,  2025, 
https://ideas.repec.org/a/ovi/oviste/vxxivy2024i2p164-168.html 
20.  Bias in artificial intelligence for medical imaging: fundamentals, detection, avoidance, mitigation, challenges, ethics, and 
prospects  -  Diagnostic  and  Interventional  Radiology,  accessed  July  30,  2025,  https://dirjournal.org/articles/bias-in-
artificial-intelligence-for-medical-imaging-fundamentals-detection-avoidance-mitigation-challenges-ethics-and-
prospects/dir.2024.242854 
21.  Ethical  considerations  for  artificial  intelligence  in  medical  imaging:  Data  collection,  development,  and  evaluation  - 
Research  Profiles  at  Washington  University  School  of  Medicine,  accessed  July  30,  2025, 
https://profiles.wustl.edu/en/publications/ethical-considerations-for-artificial-intelligence-in-medical-ima-2 
22.  AI-Guided  Ultrasound:  From  Innovation  to  Impact  |  GE  HealthCare  (India),  accessed  July  30,  2025, 
https://www.gehealthcare.in/insights/article/ai-guided-ultrasound-from-innovation-to-impact 
23.  2025  AI  Trends  for  Healthcare  and  Life  Sciences  |  Google  Cloud,  accessed  July  30,  2025, 
https://cloud.google.com/resources/ai-trends-health 
24.  2025  and  the  Next  Chapter(s)  of  AI  |  Google  Cloud  Blog,  accessed  July  30,  2025, 
https://cloud.google.com/transform/2025-and-the-next-chapters-of-ai 
25.  AI Principles - Google AI, accessed July 30, 2025, https://ai.google/principles/ 
26.  Clinical  obstacles  to  machine‑learning  POCUS  adoption  and  system‑wide  AI  implementation  (The  COMPASS‑AI 
survey),  The  Ultrasound  Journal,  published  July 3, 2025,  accessed  July 30 2025, 
https://theultrasoundjournal.springeropen.com/articles/10.1186/s13089-025-00436-2 
27.  ROI Potential with Ultrasound AI, Exo, published June 28 2024, accessed July 30, 2025, https://www.exo.inc/article/roi-
potential-with-ultrasound-ai 
28.  Application  of  Machine  Learning  to  Proteomics  Data:  Classification  and  Biomarker  Identification  in  Postgenomics 
Biology 
https://pmc.ncbi.nlm.nih.gov/articles/PMC3837439/ 
 