MedPath Project
An AI-powered oncology clinical infrastructure project designed to connect cancer patients, clinical trials, and specialist collaboration in Japan. The project aims to structure hospital data, trial protocols, DICOM and clinical documents, and create a secure, workflow-oriented platform for AI-assisted trial screening, Doctor-to-Doctor (DtoD) consultation, OCR-based data extraction, and hospital-centered interoperability.
[ TIMELINE ]
In Progress
[ CLIENT ]
NCCH
[ INDUSTRY ]
Government
Medical
[ CHALLENGE ]
In Japan’s oncology care environment, connecting the right patient to the right clinical trial remains a highly manual, fragmented, and inefficient process. Clinical trial information is distributed across multiple registries such as jRCT, UMIN-CTR, and ClinicalTrials.gov, each with different data structures, terminology conventions, update cycles, phase definitions, and recruitment status expressions. Even when trial search is technically possible, the information that actually matters in clinical practice—such as key inclusion and exclusion criteria, line of therapy, biomarker requirements, dosing schedule, visit frequency, hospitalization requirements, and operational feasibility—is often buried in long protocol documents and not readily usable at the point of care. As a result, clinicians often face a system where trials are searchable, but not easily interpretable for real-world patient matching.
A major challenge is that patient data and trial data do not naturally connect. Within hospitals, relevant oncology patient information may exist across EMRs, SS-MIX2, HL7 FHIR interfaces, PDF reports, referral letters, pathology reports, laboratory documents, radiology narratives, and DICOM metadata. However, meaningful clinical trial matching requires a structured understanding of disease stage, metastatic status, ECOG performance status, histology, biomarkers, prior treatment history, recurrence timing, line of therapy, concomitant medications, and organ function. In many institutions, assembling this information still depends heavily on manual chart review, and even evaluating a single patient can take significant time and specialist effort.
In practice, the problem is not solved by “automatic matching” alone. In Japan’s real clinical workflow, a referring physician at a partner hospital may register a case, while a specialist reviewer at a central institution such as the National Cancer Center Hospital (NCCH) evaluates the case and, when needed, discusses the patient through Doctor-to-Doctor (DtoD) online consultation. This means the platform cannot be just a search engine. It must support a workflow that connects trial discovery, case review, specialist interpretation, and collaborative decision-making within a single operational system.
Security and data governance are equally critical. In Japanese hospital environments, patient-identifiable data and raw medical records cannot be freely moved outside institutional boundaries, and each hospital may have different network policies and security constraints. In particular, central institutions and reviewers must be able to assess cases without direct access to a partner hospital’s internal systems, while identifiers and re-identification keys remain inside the hospital. This requires an architecture based on in-hospital pseudonymized data handling, role-based access control (RBAC), TLS-secured communications, two-factor authentication (2FA), and deployment patterns that align with Japanese healthcare security expectations rather than a conventional centralized SaaS model.
Another challenge is that the system must deliver explainability, not just recommendations. In oncology trial matching, clinicians do not simply want a ranked list of candidate studies. They need to understand why a trial is a candidate, which inclusion or exclusion criteria are clearly satisfied, which criteria remain uncertain, what additional documents or tests may be needed, and when specialist review or DtoD consultation should be triggered. The real challenge, therefore, is not just to apply AI to a user interface, but to build a high-trust clinical infrastructure that combines search, structured data extraction, explainable screening, collaborative review, security, and operational usability.
In this sense, MedPath is not fundamentally a trial search product. It is a complex oncology infrastructure problem: how to securely connect patient data, trial data, and specialist decision-making in a way that fits Japanese hospital workflows and can realistically be adopted at scale.
[ SOLUTION ]
MedPath is being designed not as a trial search tool, but as an AI-powered clinical operations platform that connects cancer patients, clinical trials, and specialist collaboration into a single workable system. Our approach integrates patient data acquisition, clinical trial structuring, AI-assisted screening, DtoD consultation, and security architecture into one coherent oncology infrastructure layer.
The first core component is the structuring and semantic normalization of clinical trial data. We ingest and normalize data from major Japanese registries such as jRCT and UMIN-CTR, as well as global registries such as ClinicalTrials.gov, into a shared data model that captures trial status, study phase, indication, line of therapy, biomarker requirements, geography, and key eligibility conditions. Importantly, the platform is designed to go beyond surface-level metadata. It aims to structure clinically meaningful protocol details such as inclusion and exclusion criteria, visit schedules, hospitalization requirements, treatment cycles, and operational burden so that clinicians can review trial candidates in a decision-ready format rather than reading entire protocols manually.
The second core component is a pragmatic, multi-layer strategy for patient data acquisition. Ideally, where permitted, hospitals can connect through HL7 FHIR or SS-MIX2 interfaces to enable automatic collection and refresh of approved patient data. However, because not all institutions in Japan are ready for immediate structured integration, MedPath also supports OCR-based document ingestion and manual case registration. Referral letters, pathology reports, laboratory documents, clinical notes, and other materials can be processed to extract the structured variables required for trial matching, including stage, histology, biomarkers, treatment history, ECOG performance status, metastatic status, and organ function. This hybrid ingestion strategy is essential for real-world adoption across hospitals with different levels of digital maturity.
The third design principle is a hybrid workflow of AI-assisted pre-screening combined with human specialist review. The system identifies candidate trials based on structured patient data and structured trial information, and presents not only candidate lists but also matching rationale, potentially unmet criteria, and areas requiring clarification. Final clinical judgment remains with physicians, and cases can escalate to DtoD online consultation between the referring physician and the specialist reviewer when needed. This is not a fully automated recommendation engine. It is a system where AI accelerates information synthesis and candidate narrowing, while human specialists retain responsibility for interpretation and clinical decision-making. In oncology, this hybrid model is essential for trust and adoption.
The fourth component is a security-centered architecture aligned with Japanese hospital realities. Patient identifiers and re-identification keys remain inside the hospital environment, while only pseudonymized or minimally necessary case data are shared externally for review. Central institutions and reviewers do not directly access partner hospital systems. Instead, the platform supports a hospital-sovereign data model in which cases are created by the referring institution and shared in a controlled manner through MedPath. The architecture is designed to support TLS-secured communication, RBAC, 2FA, and multiple deployment patterns such as outbound-only connectivity, depending on each hospital’s network constraints and security requirements. This makes MedPath a hospital-compatible clinical infrastructure layer rather than a generic cloud SaaS product.
The fifth core component is workflow-oriented product design. MedPath is built around the real clinical process: case registration, patient data structuring, AI pre-screening, candidate trial review, DtoD specialist consultation, and follow-up action. Different user roles—such as referring physicians, specialist reviewers, and administrative coordinators—are given distinct interfaces and permissions aligned with their responsibilities. This workflow-centered approach allows the platform to support real collaboration rather than isolated search behavior, and it creates the foundation for future expansion beyond trial matching into broader oncology care coordination.
Technically, MedPath is built on the convergence of AI, ontology, and secure healthcare infrastructure. AI is used for OCR, unstructured data extraction, patient–trial matching assistance, research summarization, and consultation support. Ontology and standardized vocabularies—such as ICD-11, SNOMED CT, ATC, RxNorm, and HGNC—provide the semantic layer that aligns patient data and trial data under a common clinical meaning structure. The security architecture ensures that all of this can function under the realistic constraints of Japanese hospitals and cancer centers.
Ultimately, MedPath is not just a search tool or a single-purpose SaaS product. It is a next-generation oncology clinical infrastructure platform designed to connect patients, trials, specialists, and hospitals across Japan. By improving trial accessibility, streamlining specialist collaboration, and strengthening inter-hospital workflows, it lays the foundation for a scalable national oncology network that can evolve far beyond clinical trial matching alone.
[ RESULTS ]
In Progress
Now
