TL;DR:
- Google Cloud and Kakao Healthcare partnership explores federated learning in healthcare and life sciences.
- Federated learning trains algorithms on decentralized data sources, ensuring privacy.
- Collaboration was showcased at Global Symposium on AI & DTx, emphasizing the potential of this approach.
- Patricia Florissi of Google Cloud explains the evolution of language learning models (LLMs) and their applications.
- Federated learning’s privacy-preserving strategy is applicable to genomics and proteomics.
- Overcoming initial concerns, federated learning enables secure and widespread implementation.
- Data fragmentation in healthcare is addressed by training models locally, with central result analysis.
- Federated learning empowers global studies of medical records and clinical trials without data movement.
- Kakao Healthcare and Google Cloud pilot project demonstrates the real-world viability of federated learning.
- Business model considerations crucial for federated learning success, linking profits to data contribution.
- Kakao Healthcare’s CEO envisions an integrated digital healthcare ecosystem through data aggregation.
Main AI News:
In the ever-evolving landscape of language learning models (LLMs), which have found significance across a multitude of sectors, including healthcare, a noteworthy collaboration has emerged between industry giants Google Cloud and Kakao Healthcare. Their partnership is centered around the groundbreaking concept of federated learning, now making inroads into the realms of healthcare and life sciences.
Federated learning, a powerful machine learning technique, revolves around training algorithms on diverse datasets sourced from distinct independent institutions. The uniqueness of this approach lies in its privacy-preserving nature, ensuring data security while harnessing its potential.
The distinguished Avison Biomedical Center of Yonsei University Medical College served as the platform for a lecture that unfolded this innovative narrative. A key event was the Global Symposium on AI & DTx, a co-hosted endeavor by Yonsei University System and Kakao Healthcare, which acted as the stage for unveiling the intricacies of this alliance.
During the discourse, Patricia Florissi, the Chief Technology Officer of Google Cloud, articulated the evolution of LLMs. These models have transitioned from their roots as algorithms capable of embedding semantic representations of individual words, enabling translations and predictions, to now orchestrating the completion of entire sentences. This transformation has given rise to influential LLMs like ChatGPT.
Drawing parallels, Florissi accentuated that the same foundational principles can be harnessed to solve intricate problems within the domain of healthcare genomics. An exemplary use case involves predicting the most plausible codon to conclude genomic sequences. In her words, “Just as a genome comprises a sequence of letters, each sequence signifies amino acids, and these amino acid sequences ultimately represent proteins. Hence, this technology can be wielded to anticipate protein folding patterns.”
Nonetheless, Florissi acknowledged that the widespread availability of such models faced initial constraints stemming from concerns regarding AI transparency, security, and even the emergence of hallucinations in LLMs. These obstacles have been meticulously addressed through the implementation of requisite policies and privacy safeguards, paving the way for their extensive deployment.
Within the context of healthcare, Florissi underscored the challenge of fragmented data scattered across numerous hospitals and geographical locations. Security concerns preclude centralizing this data in a repository. In response, she advocated a strategy where data remains distributed within regions. This involves the replication of data onto nodes within algorithms, effectively retaining data within its respective locations.
Clarifying the mechanics, Florissi elucidated, “The platform dispatches the model to the precise location of the data, enabling local training without data leaving the hospital. Subsequently, the model’s outcomes are transmitted to a central platform, where the amalgamation of model weights facilitates result analysis while upholding privacy regulations, local data norms, and overarching data protection statutes such as GDPR.”
With the advent of federated learning, monumental prospects unfold. Physicians worldwide, along with pharmaceutical companies, now possess the capacity to explore electronic medical records (EMR), study drug impacts, and delve into clinical trial outcomes on a global scale. All this is achieved without necessitating the movement of data across geographical boundaries, as eloquently outlined by Florissi.
In a testament to their pioneering spirit, Google Cloud and Kakao Healthcare embarked on a pilot project involving Kakao Healthcare’s HRS service and Google Cloud’s Vertex AI. The endeavor, initiated in June, strategically harnessed data from Severance Hospital and NCC. This initiative serves as a testament to the practicality of federated learning in real-world scenarios.
Kakao Healthcare Director Cho Lio presented a live demonstration of the company’s healthcare data research suite (HRS) during the pilot project. The focus lay on predicting the mortality rates of colon cancer patients who had undergone radical resection. The training was carried out on synthetic data. Notably, the director underscored that the federated learning pilot project yielded outcomes surpassing those of locally employed algorithms for data analysis within respective institutions.
Amidst these triumphs, Florissi imparted a note of caution regarding the pivotal role of a suitable business model in the success of federated learning. Given the costs entailed in training models with data, one plausible model involves distributing generated profits in proportion to the data contributed by individual hospitals.
Furthermore, Kakao Healthcare CEO Hwang Hee shed light on the company’s concerted efforts to integrate its four key business domains: virtual care, data empowerment, remote patient monitoring, and digital front door. The overarching vision involves aggregating healthcare data encompassing clinical, genomic, and patient-generated health data (PGHD), thus giving rise to a comprehensive digital healthcare ecosystem. Collaborations both local and international, such as those with i-SENS and Novo Nordisk, exemplify the strides made in advancing continuous glucose monitoring (CGM) for diabetes patients.
Conclusion:
The strategic collaboration between Google Cloud and Kakao Healthcare marks a significant stride in leveraging federated learning’s potential in healthcare. This innovative approach not only addresses privacy concerns but also revolutionizes how data-driven insights are gleaned from decentralized sources. The impact on the market is two-fold: it fosters a new level of collaboration while necessitating a reevaluation of business models to ensure equitable sharing of gains. The fusion of cutting-edge technology and healthcare exemplified by this partnership is poised to reshape the industry landscape, offering a glimpse into a data-driven healthcare future.
