Innovation to Impact – AI and Focused Ultrasound: Part 3
This three-part blog series explores the following: 1) the current state of artificial intelligence (AI) and machine learning (ML) in focused ultrasound; 2) the ecosystem steps that are transforming other therapies and that may accelerate the development and adoption of clinical focused ultrasound over the next few years; and 3) the rapidly developing field of generative AI and the potential impacts of widespread generative AI adoption on focused ultrasound over the coming decade.
Generative AI and its Future Impact on Focused Ultrasound
Introduction
In late 2022, the AI landscape experienced a seismic shift with the public release of OpenAI’s ChatGPT, a large language model (LLM). Other leading LLMs like Google’s Gemini, Anthropic’s Claude, and Meta’s LLaMA quickly followed, each showing the power of these models. They captured the public’s attention and intrigued the C-suites of healthcare companies, who began discussing the art of the possible. These products and the public reaction firmly put a spotlight on generative AI technology such as LLMs.
Why should those of us in healthcare be excited about generative AI, particularly given that many of us have been using AI in some way for years? Traditional clinical AI models have offered one level of insight, for instance, in spotting a suspicious lesion in a brain cancer patient’s MRI. But now, generative AI applications offer the potential to determine the lesion’s likelihood of rapid progression, alongside treatment recommendations based on both patient details and best practices derived from thousands of other cases.
While conventional AI models have proven valuable in tasks like identifying structures or anomalies in diverse data sets, generative AI and LLMs represent a paradigm shift in capabilities and potential applications:
- Enhanced image quality: Generative AI can dramatically improve image resolution and clarity, potentially reducing the need for repeat scans or more invasive imaging techniques.
- Real-time interpretation: Beyond mere detection, generative AI can provide near-instant, comprehensive analysis of images and ongoing treatments, offering detailed insights into tissue characteristics, pathologies, and intervention progress.
- Predictive diagnostics: By analyzing focused ultrasound treatment data in conjunction with patient history and vast databases of similar cases, generative AI can predict disease progression and treatment outcomes with unprecedented accuracy.
- Personalized protocols: These models can tailor procedures in real-time based on patient-specific factors and emerging findings during interventions.
- Automated reporting: Generative AI can produce detailed, natural language reports of tests, saving time for clinicians while ensuring comprehensive documentation.
- Training and education: By generating realistic focused ultrasound simulations, AI can enhance training for medical professionals, allowing them to practice on a wide range of virtual patient scenarios.
- Research acceleration: Generative AI can analyze vast data sets to identify new patterns or biomarkers, potentially leading to novel diagnostic criteria or treatment approaches.
- Multimodal integration: These systems can seamlessly integrate focused ultrasound data with imaging modalities and clinical information, providing a more holistic view of patient health.
- Guided interventions: In therapeutic ultrasound applications, generative AI can assist in real-time treatment planning and execution, optimizing energy delivery and minimizing collateral tissue damage.
- Democratizing expertise: By providing expert-level analysis, generative AI could make advanced analysis more accessible in resource-limited settings or for less experienced practitioners.
In essence, while traditional AI has enhanced specific aspects of diagnosis and treatment, generative AI promises to revolutionize the medical field—from research, diagnosis, treatment, and beyond. This leap forward could lead to more accurate diagnoses, personalized treatments, and ultimately, improved patient outcomes in focused ultrasound-based care.
We’ve written before about generative AI’s impact upon our workplace, and the fact that the Focused Ultrasound Foundation is committed to harnessing generative AI’s power to broaden its impact for every dollar spent. Today, we’ll cast our spotlight forward, addressing both the broad questions, “How will generative AI affect healthcare,” and specifically, “What does this mean for focused ultrasound adoption?”
We presume that the reader is familiar with generative AI basics, but we’ll provide brief context. Generative AI will change the way society accomplishes virtually every intellectual goal: whereas traditional AI models were purpose-built, focusing on classification, prediction, and recognition tasks, this new paradigm creates content—text, images, and more—based on large training corpuses. It will automate much “toil,” and more importantly, inform human decisions across large data sets, greatly improving patient outcomes. Recent advances in “agentic” workflows promise to give both greater autonomy and improved effectiveness to generative AI, meaning that future tools will be far more powerful than those we leverage today.
With this background, we return to the question: how will this technology impact both healthcare broadly, and focused ultrasound specifically? To answer this, we’ll start with generative AI’s audience at the most fundamental levels. Firstly, how will generative AI benefit providers, i.e., the physicians and other clinical professionals who care for patients? Secondly, what does this technology offer for each of us in the role we will inevitably play, as patients? Finally, we’ll briefly survey generative AI’s potential scope beyond natural language, then pull all the pieces together. Put on your seatbelts, and let’s examine where this journey will lead over the coming years.
Clinician Assistants
Over the next decade, clinicians will increasingly use generative AI tools as sounding boards and informed assistants. Imagine an affable deputy who, early in her career, occasionally makes mistakes, but reads more voraciously than any human ever could, and weave in observations based on the patient’s electronic health record, CT and MR scans, travel history, social determinants of health, and empirical multiomic data.
This is not hyperbole. Med-PaLM 2, from Google Research, was the first generative AI tool to reach human expert level on answering USMLE-style (United States Medical Licensing Exam) questions. Multi-modality, or the use of multiple data types to predict answers, will become pervasive. Showing its potential, Google Research has likewise released Med-Palm M: a system to synthesize and communicate information from images like chest X-rays, mammograms, and more to help doctors provide better patient care. This includes multiple modalities alongside language: dermatology, retina, radiology (3D and 2D), pathology, health records and genomics. This is not a question of “if” these tools will be used, but a question of “when.” The answer will be driven in part by regulatory considerations and the establishment of usage guidelines.
Generative AI tools are particularly promising in the field of clinical decision support. Let’s consider two examples.
Scenario 1: Physicians will use generative AI for differential diagnoses. Based on a given collection of symptoms, what are the most likely patient diagnoses? Generative AI will have the ability not just to ingest patient electronic health records (EHRs), but also to consult medical literature, clinical guidelines, epidemiological information, and other sources. For complex cases, such expanded lists will comprise a valuable secondary perspective during the patient’s workup.
Scenario 2: Arriving at a diagnosis, physicians will use these tools to identify care paths for patients. Once the diagnosis is made, how should the condition be treated? Again, generative AI will have the ability to not only draw from many different studies and papers in determining the care path, but also ingest information about genetics, epigenetics, social determinants of health and other factors to plan the right interventions.
These tools have the capability to act as proficient, “well-read” assistants to provide a second opinion through consultation. The physician will maintain the last word, but given the generative AI feedback, is now more likely to both make the right diagnosis on her first attempt, and to understand the spectrum of treatment options available to the patient.
So, what does this mean for focused ultrasound, and how will these changes affect our field? Lack of physician awareness is a major problem for focused ultrasound adoption, and clinician-facing models can help considerably, if trained properly. With LLMs up to date on recent clinical reports, their use could be a boon for focused ultrasound. As new results are published, regulatory approvals are achieved, and additional clinical trials open, such generative AI models can inform unfamiliar clinicians about therapeutic focused ultrasound and provide guidance on relevant clinical trials. Conversely, ingrained bias toward legacy treatments in LLMs—due to an outdated training corpus—could restrict physician referrals, regardless of the latest clinical advances. Fortunately, the relative velocity of new models indicates that they will continue to be trained on recent developments, including emerging therapies. In summary, use of digital physician assistants is expected to increase awareness of focused ultrasound as a therapy, with referrals and demand rising accordingly.
Patient Assistants
How might patients interact with medical generative AI systems? Patients will query these tools, beginning informally with public models. Some will choose to upload test results into public-facing agents, and eventually, with legal liabilities managed, dedicated generative AI tools for patient healthcare—fully integrated with health records—will be launched.
Consider an example.
In the coming years, a patient will be diagnosed with prostate cancer and will interact with patient-facing generative AI models to understand his condition, including his progression risk, lifestyle impacts, and possible treatment options. As we discussed above for clinicians, generative AI models may likewise point the patient toward relevant clinical trials. Such increased transparency extends into the administrative realm. Generative AI will also help him understand payer details, such as extracting policies and plain-English explanations of his account and treatment options that his insurance company will cover.
Such informed patients offer the opportunity for more meaningful exchanges between the individual and his doctor, while simultaneously creating the potential for conflict when the physician disagrees with the model. In summary, these tools will deliver new sources of information into the hands of patients, simultaneously making clinicians’ jobs both easier and harder.
What do patient-facing generative AI models mean for the field of focused ultrasound? Once again, widespread generative AI use has the potential to increase demand for focused ultrasound therapy. Lack of patient awareness is an inhibitor to adoption, which can be reduced through generative AI tools. If models are trained with recent developments and do not exhibit bias toward legacy treatments, such “patient concierge” models will empower patients and allow them to navigate their care paths with informed confidence. As mentioned above, this awareness comes with both immense gains and manageable downsides. But an empowered patient, who better understands treatment options, can have a meaningful discussion with a treating physician. Given the obvious advantages of focused ultrasound within a number of clinical areas, the net result is expected to be increased patient demand for focused ultrasound treatments.
In the example above, the patient may learn that his cancer treatment options could include active surveillance, surgery, radiation therapy, hormone therapy, cryotherapy, and focused ultrasound, with the generative AI model walking him through the relative advantages of each. More sophisticated generative AI models will be granted read access to the patient’s health records and may go further in assessing lack of calcifications in the prostate, which then highlights focused ultrasound as an attractive option. Today, a patient can perform web searches to find answers, but that process can be time-consuming and confusing for the layperson. Intelligent assistants will more clearly present options to the user. Further, inclusion of personal details—such as those found in the health record—supercharge generative AI models and will arm the patient with significant knowledge. This transparency will increase potential demand for promising therapies such as focused ultrasound.
LLM Extensions beyond Written Language
We’ve discussed generative AI in its familiar consumer context, meaning its ability to ingest large quantities of natural language text and create new textual responses. But these models excel at predicting patterns, as they’ve demonstrated by effectively solving the “protein folding” problem–predicting the protein shapes arising from amino acid sequences. What happens as these models further expand beyond human language, and will such uses have any bearing on therapies like focused ultrasound?
Life sciences examples are plentiful, and pharmaceutical development is an interesting subfield where generative AI is playing multiple roles.
- Data analysis: Generative AI models are analyzing large amounts of data from past research, including chemical properties, biological interactions, and clinical trial results.
- Potential drug compound identification: These models can then generate new molecular structures with desirable characteristics, such as high binding affinity to a target protein or minimal side effects.
- Existing drug repurposing: Generative AI is helping identify potential new uses for existing drugs.
- Drug-target interaction prediction: Generative AI is helping predict drug-target interactions.
- Individual factor considerations: Generative AI can consider an individual’s genetic makeup, lifestyle, and disease characteristics.
- Time / cost reduction: Generative AI could reduce the time and cost of developing new molecules.
Such developments can potentially have a positive impact on the focused ultrasound ecosystem, since focused ultrasound can be a major delivery system for new therapeutics. Such drug delivery mechanisms include blood-brain barrier (BBB) opening, as well as temperature sensitive drug carriers and focused ultrasound–enhanced permeability, among other mechanisms. An explosive growth in new drug innovations could result in heightened focused ultrasound interest from both pharmaceutical companies and from provider institutions to accommodate therapy deliveries. Subject to regulatory approvals, a new age of advances in the treatment of neurodegenerative diseases is ahead of us, and because these drugs must often be delivered to the brain, focused ultrasound will become an invaluable tool in therapeutic delivery.
Other Clinical and Administrative Insights
Over the coming years, generative AI will increase operational efficiencies and provide new understanding in both clinical and administrative arenas. Generative AI tools will scour our data for clues on causality, unexpected outcomes, and anomalies—performing detailed analyses that would be cost-prohibitive to manually accomplish. In turn, these developments will have both subtle and profound impacts on the healthcare landscape. Let’s consider three possibilities for improved clinical and administrative insights.
Health Economics and Outcomes Research (HEOR). Both regulatory approvals and insurance/payer considerations will benefit from LLMs and the proliferation of generative AI. Homing in on the payer aspect of this equation, today’s insurers spend considerable effort studying HEOR to generate evidence of the value of new interventions. Specifically, HEOR examines and evaluates the economic implications of treatments, diseases, and conditions, considering both direct and indirect costs, to deliver insights for healthcare decision makers.
As payer HEOR is simplified through LLM-based analysis, patients will benefit as impactful therapies receive reimbursement decisions with new efficiency. In this world of improved knowledge and insights, reimbursement of focused ultrasound should increase. If focused ultrasound provides superior treatment outcomes, then reimbursement barriers will begin to fall and focal therapy demand will rise, accordingly.
Combined therapeutic modalities. Generative AI will enable better predictions for complex multifactor treatments. For instance, in the future, rather than single modality treatments (e.g., radiotherapy OR focused ultrasound), complex interactions could be more accurately modeled and understood. This opens the door to fractional treatments across modalities, where “competing” therapies are now routinely interleaved for the best patient results.
In turn, this means that focused ultrasound will increasingly be used in complementary roles with other therapies. Despite a fractional role, focused ultrasound equipment and expertise will be needed. Our collective therapeutic future is not “X or Y,” but “X and Y, together in optimal combination,” and the focused ultrasound ecosystem will see demand rise accordingly.
The rising tide and patient preferences. As is often stated, a rising tide lifts all boats. With better treatment assessment, enhanced image quality, and other improvements enabled by generative AI, virtually all treatment modalities will improve, driven by both automated analyses and imaging improvements.
For many significant conditions, past treatment options were very limited, e.g., surgical removal of a lesion. As clinical advances have expanded options, clinicians and patients have increasingly considered numerous factors to chart a treatment course. Continued improvements—offering multiple promising options for a given patient—mean that therapies will be chosen based increasingly on economic factors and patient preferences. The fact that focused ultrasound is noninvasive may lead it to become a preferred option over other more disruptive therapies. A patient’s ability to choose a treatment requiring no incisions nor radiotherapy effects may lead to increased demand for focused ultrasound.
Tying it all together
In the examples above, we’ve only scratched the surface of the many advances that generative AI will enable. Clinical language models, advances in basic life sciences understanding, and numerous other avenues will be opened, with patients benefiting universally. The potential of generative AI with focused ultrasound is creating a groundbreaking new paradigm for healthcare. Today, it is possible to envision a highly personalized form of noninvasive healthcare much earlier than many experts predicted, bringing great hope for future patients.
Generative AI will help us redefine our own personal health journeys. In normal times, when we have no obvious pathologies, generative AI will help us maintain our health. When a problem arises, generative AI will help us pinpoint the right diagnosis. And with that diagnosis in hand, generative AI will allow us to identify the optimal treatment path based on our personal data as well as empirical evidence across the population. All these advantages will likewise be layered atop an improved strata of business services, accomplishing more at reduced cost through reduction/elimination of unnecessary administrative expenses. For example, generative AI will help find anomalies which point to fraud, such as detecting the provider who bills for unperformed procedures or “up-coding” medical procedures to get higher payment. By automating auditing processes for medical coding and billing, it will reduce overall healthcare costs.
Generative AI will open doors of understanding and shine light into dark corners within the medical field. It’s an exciting time to be not just watching but driving this revolution in healthcare. Focused ultrasound, and the patients who will receive treatment, are poised to be beneficiaries of this revolution, for all the reasons outlined here. We must be aware of this potential, stay cognizant of the opportunities, and work as always to push the boundaries of clinical care.
Rick Hamilton is the Foundation’s managing director and chief technology officer. He is recognized as one of the most prolific inventors in world history, with over 1,060 issued US patents.
Kerrie Holley is a member of the National Academy of Engineering and author of the books, “AI-First Healthcare” and “LLMs and Generative AI for Healthcare,” among others.
Read the Blog Series
Part 1: Focused Ultrasound + AI/ML Today: The Intersection of Two Rapidly Evolving Fields
Part 2: Focused Ultrasound + AI/ML Tomorrow: Maturing Preclinical and Clinical Capabilities
Part 3: Generative AI and its Future Impact on Focused Ultrasound