Our Platforms

The AI Centre are developing a range of technologies and platforms, to enable better identification of diseases early, more accurate diagnosis and personalised treatment.

These Platforms to Support Full Lifecycle AI Development, taking models from conception, to development and deployment, within the clinical environment.

The following technologies are currently under development:

• FLIP – Federated Learning Interoperability Platform: FLIP links data from multiple NHS Trusts to enable AI at scale. It includes secure data storage for processing and analysis within each of our partner NHS Trusts – a secure enclave or area within the firewall that keeps sensitive patient data inside the Trust. Our federated learning approach brings algorithms to the data within each NHS Trust’s secure enclave, without needing to share information outside the secure firewall or break local governance rules. Algorithmic models are sent to multiple Trusts and trained on local data before being securely combined to achieve consensus. The model is then applied within each secure enclave, where it learns from the data, is updated again, and the process repeated until an improved consensus model is created. To achieve convergence, the process of learning and combining is reiterated, until each locally applied model reaches the same conclusion, indicating that the model is generalizable and can be consistently applied.

• OncoLlama is an open-source, NHS developed, Language AI tool that reads cancer patients' medical letters and reports, automatically extracting detailed information including cancer type, stage, genetic markers, and treatment responses with >98.5% accuracy. The
  system transforms millions of unstructured clinical documents into searchable,
  analysable data. This can enable powerful research, improved cancer care
  planning, and help to plan for and accelerate clinical trials. The AI model is
  deployed entirely within NHS firewalls, removing the need for sensitive data to
  be sent to third parties.

• Haystack: a tool to automate clinical trial recruitment, leveraging large language models and both structured and unstructured clinical data to dynamically match patients to suitable clinical trials in real time. Using Large Language Models (Fine-Tuned Llama 3.1 8B LLM- OncoLlama) for Complex Cancer Data Extraction from Free Text to extracts Cancer Facts including: morphology, topography, biomarkers/variants, pathological findings, metastases, etc; Treatment and Event Timeline including response and toxicities; Patient Facts including performance and PROMs; and New Treatment Changes. We now have both deep and near live visibility over very detailed phenotypes, that helps identify patients for clinical trials with a deal of precision that's not been possible before.

• GeNotes: commissioned by Genomics England – GeNotes aims to improve the accessibility and usability of genomic clinical guidelines for clinicians. To enhance this capability, we are developing a Retrieval Augmented Generation (RAG) based conversational chatbot. This chatbot will allow healthcare professionals to efficiently retrieve relevant genomic clinical guidelines from the GeNotes dataset.
• AiSeptron: NIHR and Evelina Charity funded study designing a paediatric sepsis prediction tool using machine learning, for which the AIC is co-ordinating data delivery, infrastructure and supporting development for this cross-site, prospective observational study across 6 different hospitals over 18 months.
• Work as part of the Genomic AI Network: Developing Large Language Models for high precision medical concept extraction from unstructured text reports, streamlining genomic testing and improving patient identification for clinical trials. The technology developed and validated at GSTT will be made available with support for its adoption by NHS Trusts nationwide.
  • GenoLlama is fine tuned to extract biomarker information from genomic test result reports, with potential in improved recruitment into clinical trials leading to better patient outcomes. This work may also be used to retrospectively populate the Unified Genomics Record, in collaboration with the Genomics Unit at NHS England; as well as to speed up data processing work at the National Disease Registration Service.
  • PhenoLlama is fine tuned to identify and codify symptom information as terms in the Human Phenotype Ontology for improved referral to genomic testing, particularly in patients with suspected inherited cardiac conditions.

Video demonstration of OncoLlama, PhenoLlama and GenoLlama at the Clinical AI Interest Group – of NHS Genomic AI Network

•  Cogstack:
  This ‘in-house’ spin out from KCL, KCH, UCLH, SLaM and GSTT had been operating
  in our hospitals for a number of years. Its platform uses Natural Language
  Processing (NLP) to extract concepts from medical records and Machine Learning
  models (MedCat) to speed up diagnosis, treatment options, enhance coding and
  improve data quality.

•   DeepC
  DeepC is an external company that provide an AI deployment platform which synchronises radiology and radiotherapy applications in a wide range of clinical pathways. It orchestrates data flows and predictions, and provides platform for monitoring, across radiology AI applications. The AI Centre and DeepC entered a collaboration agreement on use and future developments of the platform. NHS trusts are also using the DeepC platform to deploy in-house AI Models in radiology and radiotherapy, such as AutoSeg: a chemotherapy automated organ contouring application.