Over the last few years, the use of artificial intelligence models to improve healthcare delivery has been gaining momentum. While models like IBM’s Watson Health (now Merative) and GPT 4 Med have been available for a few years now, access was restricted to paying, enterprise customers. With Google making its MedGemma 27B and MedSigLIP models open-source and available to everyone, the conversation around generative AI in healthcare has now firmly shifted from access to effective deployment. These powerful, open-source, multimodal models can process medical text, images and speech almost instantly and with extremely high levels of accuracy. A clinician can input an image, say an X-ray, into the system and get a diagnosis as well as the recommended line of treatment within minutes. In principle, this can revolutionise healthcare by making it possible for healthcare professionals anywhere in the world to access and gain from these models. The reality is a little more nuanced! 

Given the intricacies of healthcare and health tech, a simple plug-and-play model of AI deployment will not be feasible for the foreseeable future. Open-source models are based on wide swathes of generalised data; specificity is the key to getting healthcare right. Knowing which model to deploy and within what framework is critical for an industry like healthcare, which is highly regulated, and where the smallest error can have far-reaching consequences in the real world. Unlike pure-play technology deployments, AI in healthcare is not simply about plugging the model into an existing system and going live with it. Knowing where in the clinical workflow these models need to be integrated is crucial to maximising their effectiveness.

Open-source models like MedGemma 27B can be deployed in areas that require free text generation, like drafting radiology reports or clinical note summaries, while MedSigLIP is used for multimodal imaging tasks like classification and retrieval. However, deploying these models in a raw format lacks the clinical perspective and nuance required in real-world situations. An X-ray analysed by a generic model is likely to offer more general observations as compared to the same image being analysed by a system that has regional epidemiology and regulatory protocols built into it. This additional information can provide more relevant, actionable insights, which can impact clinical decision-making and ultimately patient outcomes. 

Being able to do this effectively requires more than technical skills; it needs deep domain knowledge and an understanding of the operational realities of aspects like pathology integration, radiology reporting, ICU triage and chronic care workflows. Equally important is an understanding of how healthcare functions in the real world: how actual clinicians work and how hospitals and clinics function within their local regulatory frameworks and clinical specialities. Without this context, even the most advanced AI model stands the risk of being ineffective or underused in real-world settings, or in the worst case, wrong, and therefore quite damaging. 

Different countries have specific regulatory frameworks which healthcare providers must adhere to. Any tech intervention, like an AI tool or model, must meet those guidelines. Leading health tech firms already have internal guidelines and frameworks around GenAI Quality and Trust, which are designed to bring rigour to their AI deployments. These guidelines address critical issues like bias mitigation, explainability and safety checks. For instance, doctors and clinicians using these tools also need to be able to question the system and understand the thinking behind its recommendations in real time.  

Companies that already work with healthcare customers come with an understanding of domain-specific intricacies and challenges, and stand to have a clear advantage over large technology companies, where the focus would primarily be on the AI aspect. Several Indian companies have a strong presence in this space, servicing providers and payers in other geographies. Combining their proprietary knowledge of healthcare systems with open-source AI models can have a significant impact on large numbers of patients and healthcare professionals. 

Going Beyond Technology  

Open-source AI needs domain-specific talent to truly make an impact in a field like healthcare. This is where India has a potential advantage that it must capitalise on. The country’s prowess in engineering and technology talent is well established globally. Indian companies must seize this advantage and ensure that they go beyond just broad generative AI training and focus on building deep domain knowledge among their workforce. India is home to a significant number of health tech firms, both Indian and global enterprises, that have deep domain skills in healthcare. As these companies build niche AI skills on top of the functional depth they have, it can pave the way for India to become a powerhouse in AI in healthcare.
These teams are already conversant with AI frameworks, and also the specific language of healthcare, whether it is DICOM protocols, FHIR schemas or EMR logic or compliance frameworks like HIPAA and GDPR. This inherent and contextual knowledge makes it possible for them to easily deploy AI models in a manner that will quickly impact patient outcomes. The country’s vast medical infrastructure could serve as a test bed for building up AI models that can then be scaled globally. 

Also read: Transforming Healthcare Research in India with AI-Driven Microscopy and Next-Gen Diagnostics

For Indian enterprises, upskilling the existing workforce is also the best approach to building a long-term, sustainable skill set in an industry that is rapidly evolving. With skills and knowledge becoming redundant every few months, building upon the existing knowledge framework can help mitigate the likelihood of your workforce becoming redundant in a rapidly changing world. The future of AI in healthcare will be shaped by domain-aware digital architects who understand both algorithms and anatomy. As open-source models like MedGemma 27B and MedSigLIP become widely available, it’s these domain-anchored players, strong in clinical logic and engineering execution, who will be best equipped to deploy AI safely and ethically into patient-facing environments. If Indian firms capitalise on their existing expertise in this field and build AI native expertise on top of this, then India will be all set to shape the future of global healthcare. The countries and companies that get this equation right—open models × regulated workflows × domain-trained talent—will shape the future of GenAI in healthcare.

Views expressed by: Sumit Bhardwaj, EVP and Head of Operational Excellence, CitiusTech

Like

Dislike

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter , Instagram.

Disclaimer: The views and opinions expressed in this article are solely those of the author and do not necessarily reflect the official policy or views of any organisation. The content is intended for informational and educational purposes only and should not be construed as medical advice.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!