LizAI XT -- Artificial Intelligence-Powered Platform for Healthcare Data Management: A Study on Clinical Data Mega-Structure, Semantic Search, and Insights of Sixteen Diseases

TL;DR

The paper tackles the challenge of fragmented and privacy-sensitive healthcare data impeding AI-driven clinical insights by introducing LizAI XT, an AI-powered platform that mega-structures diverse clinical data into disease-specific, anonymized datasets. It presents an end-to-end pipeline combining NLP, computer vision, speech, and multimodal processing with ontology-driven knowledge graphs to ensure standardization and interoperability (FHIR, HL7, DICOM) and uses embedding-based representations for accurate variable extraction. On a synthetically generated, clinically grounded dataset of 16,000 patients, 16 diseases, ~112,711 files, and ~781 variables, LizAI XT achieves an overall accuracy of $95.79\% \pm 5.69\%$, with exceptionally high performance in colorectal cancer ($99.12\% \pm 0.049\%$), prostate cancer ($99.03\% \pm 0.08\%$), COPD ($98.89\% \pm 0.076\%$), and contraception ($98.28\% \pm 0.12\%$). Retrieval speed is sub-second per variable per patient on a minimal 4x NVIDIA A30 GPU cluster, and the system supports on-premises or cloud deployment with strict privacy controls, enabling scalable, client-governed healthcare data management. The authors position LizAI XT as a complementary platform to EMR/EHR and cloud solutions, capable of powering real-time analytics, precision medicine, and nationwide data initiatives while maintaining regulatory compliance and data security.

Abstract

AI-powered LizAI XT ensures real-time and accurate mega-structure of different clinical datasets and largely inaccessible and fragmented sources, into one comprehensive table or any designated forms, based on diseases, clinical variables, and/or other defined parameters. We evaluate the platform's performance on a cluster of 4x NVIDIA A30 GPU 24GB, with 16 diseases -- from deathly cancer and COPD, to conventional ones -- ear infections, including a total 16,000 patients, $\sim$115,000 medical files, and $\sim$800 clinical variables. LizAI XT structures data from thousands of files into sets of variables for each disease in one file, achieving >95.0% overall accuracy, while providing exceptional outputs in complicated cases of cancers (99.1%), COPD (98.89%), and asthma (98.12%), without model-overfitting. Data retrieval is sub-second for a variable per patient with a minimal GPU power, which can significantly be improved on more powerful GPUs. LizAI XT uniquely enables fully client-controlled data, complying with strict data security and privacy regulations per region/nation. Our advances complement the existing EMR/EHR, AWS HealthLake, and Google Vertex AI platforms, for healthcare data management and AI development, with large-scalability and expansion at any levels of HMOs, clinics, pharma, and government.

Figures (7)

• Figure 1: Overall illustration of a fully secured AI-powered LizAI XT Platform for client-controlled healthcare data management. This report focuses on the data management by clinical data mega-structure using LizAI XT.
• Figure 2: Some examples of our generated medical files for a patient: (A) a spreadsheet containing key clinical notes from doctors, (B) a scanned PDF of a checklist, (C) a printed PDF report summarizing the patient's complete medical history, and (D) a PDF report of laboratory test results, covering macroscopic, microscopic, and chemical examinations.
• Figure 3: System diagram of Clinical Data Mega-Structure by LizAI XT which automates the data collection, anonymization, storage, and structuring of medical data from various healthcare systems. The system de-identifies personal data (1), stores it in a high-performance object storage system (2), and routes it to specialized processing components based on data type (3). NLP, computer vision, speech processing, and multimodal analysis (4) are enhanced by ontology and knowledge graphs (5) for improved accuracy. Processed data undergoes refinement (6) before clinical variables are extracted (7) and structured into disease-specific datasets for research, analytics, and decision support (8).
• Figure 4: Illustration of data mega-structure by LizAI XT. Thousands of files and information of 16,000 patients in this study are fragmented in different types and formats, which can efficiently be structured into one data-table per disease by relevant variables.
• Figure 5: Assessment of LizAI XT’s performance based on accuracy of data mega-structure. The accuracy is calculated for each disease, and the overall performance is calculated for structuring the entire database.