Back to Compliance

Regulatory Documentation

Technical documentation and compliance information for Apolo Medical Model

MDR 2017/745 CE Class IIa (Pending) FDA 510(k) (Planned) ISO 13485:2016 ISO 14971:2019 IEC 62304:2006

Regulatory Status

Apolo Medical Multimodal Instruct is classified as a Software-as-a-Medical-Device (SaMD) that falls under the following regulatory frameworks:

EU Medical Device Regulation (MDR) 2017/745

Classification: Class IIa (Rule 11)

Status: Technical documentation complete, CE mark pending Notified Body review

FDA (US)

Classification: Class II medical device

Status: 510(k) submission planned for Q3 2025

Key Device Information

Device trade name	APOLO Medical Multimodal Instruct
Basic-UDI-DI	TBD (to be assigned after NB engagement)
Intended purpose	Software-as-a-Medical-Device (SaMD) that assists qualified healthcare professionals by converting anonymised medical images into structured visual observations (Stage 1) and by providing explanatory diagnostic reasoning (Stage 2).
SaMD category	IMDRF SaMD Class II (Inform clinical management) / MDR Rule 11 – Class IIa
Operating environment	On-premise hospital servers or secure clinical workstations (Windows 10/11, Ubuntu 20.04+, NVIDIA GPU w/ ≥ 8 GB VRAM)
Supported modalities	Radiology (X-ray, CT, MRI), Ophthalmology (Fundus, OCT), Pathology, Dermatology

MDR Technical Documentation

In accordance with the Medical Device Regulation (EU) 2017/745 Annex II, Apolo maintains comprehensive technical documentation to demonstrate conformity with the applicable requirements.

1. Device Description and Specification (Annex II §1)

Functional Specifications:

Stage 1 (Apolo-VL): Accepts DICOM/JPEG/PNG images; outputs structured JSON + Markdown description.
Stage 2 (Apolo-Instruct): Accepts Stage 1 description; outputs <think> reasoning + conclusion.
Unified Multimodal Path: Optional direct inference for research sites (not in clinical release).

2. Design and Manufacturing Information (Annex II §2)

Software Architecture

┌──────────────────────────┐ ┌──────────────────────────┐ │ Staging PACS Listener │ → │ Stage 1 – Apolo‑VL │ └──────────────────────────┘ │ (Docker container) │ └───────────────┬──────────┘ │ JSON ↓ ┌──────────────────────────┐ ┌───────────────┴──────────┐ │ Stage 2 – Apolo‑Instruct │ ← │ Secure Message Broker │ │ (Docker container) │ └──────────────────────────┘ └───────────────┬──────────┘ ↓ HL7 /FHIR Output

Key aspects of the design and manufacturing process:

Containerisation: Each stage packaged as Docker image; SHA-256 digests tracked in GitLab registry.
CI/CD: GitLab-CI pipeline with unit tests, integration smoke tests, SBOM generation.
Configuration control: Git tag → build artifact mapping; IEC 62304 state = "SOUP with controlled updates".

3. Verification and Validation

Apolo undergoes rigorous verification and validation to ensure it meets its intended purpose:

Software verification according to IEC 62304 practices
Non-clinical performance testing on standard datasets
Clinical performance evaluation in real-world settings
Usability validation with healthcare professionals

4. Post-Market Surveillance (PMS) Plan (Annex II §5)

Apolo's PMS plan includes:

Monitoring diagnostic performance drift every 3 months
Capturing user feedback, adverse events, and near-misses
Quarterly PMS review meetings
Trending of accuracy metrics vs. baseline
CAPA initiation if AUC drop > 0.05 or user-reported harm
Periodic Safety Update Report (PSUR) every 2 years

Risk Management (ISO 14971)

Apolo's risk management process follows ISO 14971:2019 methodology, with a comprehensive risk management file that documents the analysis, evaluation, control, and monitoring of risks associated with the device.

Risk Management Process

Systematic identification of hazards and hazardous situations
Estimation of risks for each hazardous situation
Risk evaluation against acceptability criteria
Implementation and verification of risk control measures
Evaluation of residual risk acceptability
Ongoing risk monitoring during post-market surveillance

Top-Level Risk Table (excerpt)

Hazard	Foreseeable sequence	Harm	Initial RPN	Mitigation	Residual RPN
Misdiagnosis due to incorrect reasoning	Incorrect Stage 1 description → misleading Stage 2 output	Delayed treatment	12	Mandatory clinician review; explanation trace; confidence score	4
Data breach	Raw image leaked outside secure zone	Patient privacy violation	10	On-prem deployment; TLS; no cloud egress	3
Model drift	Performance degradation over time	Diagnostic inaccuracy	9	Scheduled re-validation & monitoring dashboard	4

Risk-Benefit Analysis

The residual risks associated with Apolo have been carefully analyzed and determined to be acceptable when weighed against the benefits of improved workflow efficiency and diagnostic support, provided that proper clinician oversight is maintained.

Key Risk Control Measures

Explicit system design requirement that AI output is advisory only, with mandatory human oversight
Clear labeling and user interface elements that distinguish AI-generated content
Transparency through explanatory <think> tags that expose the reasoning process
Confidence scoring to indicate uncertainty levels
Two-stage architecture that allows for cross-verification
Robust validation on diverse datasets to minimize bias

Performance Data

Apolo has undergone extensive performance evaluation to validate its safety and effectiveness for its intended purpose. The evaluation includes both analytical and clinical performance assessments.

Analytical Performance

Testing was conducted on well-established, publicly available datasets:

MIMIC-CXR (v2.0)
EyePACS
CheXpert
ROCO
OCTA-500

Performance Metrics

Task	AUC / Score	95% CI
DR detection (EyePACS ≥ moderate)	0.94	0.93–0.95
AMD detection (AREDS)	0.92	0.90–0.93
CheXpert 5-label average	0.92	0.91–0.93
ROUGE-L (description vs refs)	0.49	0.48–0.50

Clinical Performance

A prospective pilot study was conducted at a clinical site with 120 cases, demonstrating:

Mean time-to-report reduced by 18%
No critical misses compared to senior radiologist ground truth
High user satisfaction ratings (mean 4.6/5 on explanation clarity)

Ongoing Clinical Evaluation

As part of our post-market clinical follow-up (PMCF) plan, Apolo will undergo additional multicentre evaluation (n ≥ 500) over a 12-month period to capture edge cases and rare findings. This proactive approach ensures continuous validation of the system's performance in diverse real-world settings.

Performance Limitations

Users should be aware of the following limitations:

Apolo is designed to assist, not replace, clinical judgment
Performance may vary across different imaging equipment, protocols, and patient populations
The system has not been validated for all possible pathologies and should be used within its validated scope
Stage 1 description quality directly impacts Stage 2 reasoning accuracy

Technical Documentation Downloads

The following technical documentation is available for download:

MDR & CDS Conformity Report

Comprehensive assessment of Apolo's conformity with MDR requirements and Clinical Decision Support validation

Apolo