DEMO MODE — Pre-computed ResNet-50 inference results. No live backend.
You are offline. Symptom screening works fully offline. New diagnoses will queue and sync when reconnected.
Home/Project Documentation
CSC 492 · Academic Project Record

Project Documentation

System analysis, design, implementation notes, and evaluation plan for MediSecure AI. This record distinguishes what is visible in the current codebase from what remains proposed or requires verification.

IMPLEMENTEDPARTIALLY IMPLEMENTEDPROPOSEDPENDING VERIFICATION

Developed by Anyoha Elvis Chidumebi, Computer Science Student · Supervisor: Dr. Ugo Nwankwo · HOD: Prof. Arinze Nwaeze

Foundation

Project Overview

MediSecure AI is an existing Vite, React, and TypeScript web application for demonstrating a privacy-aware medical-image decision-support workflow in a Nigerian academic context. The current repository contains public pages, role-oriented dashboards, Blink database access wrappers, client-side privacy utilities, offline/PWA mechanisms, and a demo inference mode.

It is important to separate the working interface from the complete research system described in some product copy. The inspected repository does not contain model weights, a training service, a federated coordinator, a custom backend, or automated behavioural tests. Those areas are labelled PARTIALLY IMPLEMENTED, PROPOSED, or PENDING VERIFICATION below.

Student
Anyoha Elvis Chidumebi
Academic context
CSC 492 · Caritas University
Supervisor
Dr. Ugo Nwankwo
Application status
Research prototype / decision-support demonstration

Academic integrity rule: claims about hospitals, patients, model performance, certifications, deployments, security controls, and clinical usefulness are not treated as implemented unless they can be verified from code or supplied evidence.

Chapter 1

Introduction

1.1 Background

Medical image analysis can support clinicians by highlighting patterns for further review. Health data is sensitive, institutions may not be able to centralise patient images, and a research prototype must not be presented as a certified medical device. MediSecure AI explores a privacy-aware interface around chest X-ray classification and federated-learning concepts.

1.2 Motivation

The motivation is to explore how institutions could collaborate on machine-learning research while retaining local control over source data. This is an academic and architectural motivation; the repository does not prove a live hospital deployment or a completed clinical study.

1.3 Objectives

  • Provide a role-oriented web interface for a medical AI research prototype.
  • Represent users, cases, screenings, hospital nodes, training rounds, privacy settings, models, and audit events.
  • Demonstrate client-side privacy utilities such as PII stripping, noise mechanisms, and RDP-style accounting.
  • Provide explicit consent, medical-disclaimer, and research-prototype messaging.
  • Identify the implementation gaps that must be addressed before real deployment.

1.4 Methodology Summary

The most defensible description of the visible development approach is iterative Agile-style development: requirements, planning, design, development, testing, review, refinement, deployment, and maintenance. Formal sprint records were not found in the repository.

1.5 Results / Expected Results

DOCUMENTED The current result is a working research-oriented web interface with demo-mode flows and typed data access. Model performance, clinical usefulness, federated training performance, and privacy guarantees are not verified results in this repository.

1.6 Limitations and Scope

The verified scope is the existing frontend, Blink SDK data access, UI-level role gating, client-side privacy helpers, legal/safety pages, and offline/PWA mechanisms. Training, inference infrastructure, secure node transport, server enforcement, and automated behavioural testing remain incomplete or unverified.

Chapter 2

Literature Review

2.1 Introduction

This review is structured around the literature themes named by the existing About page. The abbreviated references below require bibliographic verification before inclusion in a final thesis.

Author / yearStudy / technologyMethod / key relevanceLimitation for this project
Abadi et al. (2016)Deep learning with differential privacyDP-SGD backgroundFull citation and exact application require verification
McMahan et al. (2017)Communication-efficient learning from decentralised dataFedAvg backgroundNo executable aggregator in current repository
Kaissis et al. (2020)Privacy-preserving federated learning in medical imagingHealthcare privacy contextExternal source must be checked
Sheller et al. (2020)Federated learning in medicineMulti-institution collaboration contextExternal source must be checked
Beaulieu-Jones et al. (2019)Privacy-preserving distributed deep learningDistributed learning comparisonExternal source must be checked
Rieke et al. (2020)Future of digital health with federated learningDigital health contextExternal source must be checked
Kermany et al. (2018)Medical image classification dataset / deep learningResearch dataset contextDataset provenance and license require verification

2.2 Artificial Intelligence in Medical Diagnosis

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.3 Machine Learning in Healthcare

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.4 Deep Learning

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.5 Medical Image Classification

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.6 Chest X-ray Analysis

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.7 ResNet-50

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.8 Federated Learning

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.9 FedAvg

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.10 Differential Privacy

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.11 DP-SGD

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.12 Privacy-Preserving Machine Learning

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.13 Secure Communication in Healthcare

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.14 Healthcare Data Protection

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.15 Nigerian Data Protection Context

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.16 Existing Related Systems

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.17 Comparative Analysis

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.18 Research Gap

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.19 Conceptual Framework

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

2.20 Chapter Summary

This required subsection should be expanded using verified scholarly sources. In the current application, the related concept is represented in UI copy, domain types, or client utilities rather than a complete research implementation. Mark claims about performance, novelty, deployment, regulation, or clinical usefulness as requiring verification.

Research gap — defensible wording
The repository offers a unified educational interface that brings together role-based workflows, federated-learning terminology, privacy utilities, audit views, and medical safety messaging. The repository does not establish that it is the first system, clinically validated, or deployed in hospitals.
Chapter 3

Methodology & System Analysis

3.1 Software Development Methodology

Agile Software Development is suitable because the project combines changing interface requirements, AI experimentation, continuous testing, security refinement, and incremental implementation. Scrum should not be claimed unless separate Scrum evidence exists.

StageActivitiesDeliverablesAdvantages / limitations
RequirementsIdentify roles, workflows, privacy and safety needsRequirement listKeeps scope visible; may change during research
PlanningPrioritise increments and risksIteration planSupports incremental delivery; requires discipline
DesignDesign routes, entities, diagrams and responsive UIDesign specificationFinds inconsistencies early
DevelopmentImplement React modules, SDK wrappers and utilitiesWorking incrementsFast feedback; partial features must be labelled
TestingRun static checks and planned behavioural testsTest evidenceCurrent behavioural suite is pending
Review / refinementInspect preview, claims, links and security boundariesRefined releaseReduces unsupported claims
Deployment / maintenancePublish, monitor, update and documentPublished prototypeReal deployment requires governance

3.2 Existing System Analysis

The existing system analysed here is a conceptual centralised medical-data workflow, not a named hospital deployment. In such a workflow, patient images and clinical metadata may be transferred to a shared server for training or analysis. Potential weaknesses include central exposure, data ownership conflicts, institutional barriers, access-control complexity, and regulatory risk.

3.3 Proposed High-Level Model

The proposed model keeps source images at a local node, performs local model operations, applies a privacy mechanism to an update, and sends an intended protected update to a coordinator for aggregation. In the current repository, these stages are represented by domain fields and administrative screens; executable node/coordinator services are proposed or pending verification.

System Architecture
Clinician / patientMediSecure interfaceApplication + Blink data layerLocal node / model boundaryProtected update / coordinator
Context Diagram
UserMediSecure AIBlink databaseOptional inference APIAudit record
Existing System DFD
Hospital userCentral uploadCentral data storeCentral model servicePrediction / report
Proposed System DFD
User inputLocal preprocessingLocal modelProtected updateFederated coordinatorGlobal model metadata
Use Case Diagram
AuthenticateRun screeningReview reportManage nodesReview audit logsView privacy metrics
Activity Diagram
Access systemValidate credentialsVerify roleValidate inputGenerate decision-support resultClinician review
Sequence Diagram
User → InterfaceInterface → ApplicationApplication → Data layerApplication → Inference boundaryApplication → Audit log
Class Diagram
AppUserMedicalCaseSymptomScreeningFederatedNodeTrainingRoundPrivacyConfig
Component Diagram
PublicLayoutRoleLayoutRoleSidebarPage modulesBlink SDK wrappersPrivacy utilities
Entity Relationship Diagram
userscasessymptom_screeningsfederated_nodestraining_roundsaudit_logs
Deployment Diagram
Browser / PWAPublished web hostBlink SDK gatewayBlink databaseOptional external inference service
Security Architecture
Role gateConsent / disclaimerInput validationPII minimisationAudit loggingServer controls — verify
Network Architecture
Client browserHTTPS originBlink APIsOptional inference APINode transport — proposed
External entities
Clinician, patient, administrator, auditor, data manager
Processes
Authentication, screening, case/report view, privacy metrics, audit review
Data stores
Blink tables: users, cases, screenings, nodes, rounds, models, privacy, audits

3.4 Analysis and Justification

The proposed architecture is justified as a privacy-aware research direction, not as proof that the current frontend executes federated training. The distinction allows the academic report to describe design intent while remaining consistent with the codebase.

Chapter 4

System Design

4.1 Introduction

The design describes the current user interface and data model while identifying proposed backend and machine-learning components. All diagram labels use the same terminology: user, interface, application layer, local node, ResNet-50, DP-SGD, protected update, coordinator, FedAvg, and global model.

4.2 Objectives of the System Design

The design prioritises role separation, responsive access, explicit medical safety language, privacy-aware data presentation, auditability, and a clear boundary between demo outputs and verified results.

4.3 Control Centre Design

PARTIALLY IMPLEMENTED The admin area contains dashboard, federated training, nodes, models, privacy, users, audit, and settings pages. A complete operational control centre with live orchestration, certificate management, transport monitoring, and real-time alerts is PROPOSED.

Dashboard
Node status, round and model metadata where records exist
Hospital nodes
Node identity, status, dataset and TLS metadata fields
Federated training
Round, quorum, privacy and aggregation metadata
Privacy / security
Privacy configuration and audit-related screens
Users
Role, activation and approval management
Reports
Case and patient-facing report views

4.4 Interface Specifications

InterfacePurposeUsersInputsOutputsStatus
LoginAccess the applicationAll rolesEmail and roleSession or errorPARTIALLY IMPLEMENTED
Patient screeningCollect symptom responsesPatient / providerQuestion responsesRisk level / recommendationIMPLEMENTED
Medical imageCapture or select an imageProviderCamera frame / imagePreview blob / data URLPARTIALLY IMPLEMENTED
Prediction resultDisplay decision-support outputProvider / patientCase/result recordLabel, confidence, summariesPARTIALLY IMPLEMENTED
Audit logsReview platform eventsAuditor / adminSearch and severity filtersRead-only event listIMPLEMENTED
DocumentationExplain status and designAll usersNavigationAcademic chapters and diagramsIMPLEMENTED

4.5 Data Dictionary — Verified Application Tables

TableKey fieldsPurposeStatus
usersid, email, full_name, role, is_active, consent fieldsUsers and rolesIMPLEMENTED
federated_nodesid, name, institution, status, dataset_size, privacy / TLS metadataDisplayed node recordsIMPLEMENTED
training_roundsid, round_number, status, quorum, metrics, epsilonRound metadataIMPLEMENTED
privacy_configsid, config_name, epsilon, delta, noise_multiplierPrivacy configurationsIMPLEMENTED
global_modelsid, version, architecture, metrics, model_hashModel registry metadataIMPLEMENTED
casesid, provider_id, patient_id, diagnosis, confidence, findingsCase/report recordsIMPLEMENTED
symptom_screeningsid, user_id, responses, score, risk_levelScreening recordsIMPLEMENTED
audit_logsid, timestamp, event_type, severity, detailsAudit eventsIMPLEMENTED
notificationsid, user_id, message, type, is_readNotificationsIMPLEMENTED
contact_messagesid, name, email, subject, messageContact submissionsIMPLEMENTED

4.6 Algorithms and Mathematical Specifications

DOCUMENTED The source contains Gaussian and Laplace noise helpers, RDP curve and conversion helpers, PII stripping, age binning, DP count release, and budget tracking. A production DP-SGD training loop, executable FedAvg coordinator, and model evaluation pipeline are not present.

FedAvg — proposed specification
w(t+1) = Σₖ (nₖ / n) · wₖ(t+1)
wₖ is a local model update; nₖ is the local sample count; n is the total participating sample count. This is a design formula, not evidence of an executable aggregator.

4.7 Variables Used in Programming

VariableModule / sourcePurposeStatus
roleauth / routesSelected or stored user roleIMPLEMENTED
diagnosis / confidencecase and demo typesDecision-support result fieldsPARTIALLY IMPLEMENTED
epsilon / deltadp.ts and privacy typesPrivacy accounting parametersPARTIALLY IMPLEMENTED
roundNumber / quorumtraining typesRound progress metadataIMPLEMENTED
globalModel / local updatedesign terminologyFederated model conceptsPROPOSED
Chapter 5

Software Testing & Evaluation

TypeScript and ESLint are available as static checks. No dedicated unit, integration, end-to-end, model-evaluation, federated-training, privacy-accountant, performance, or user-acceptance test suite was found. Behavioural outcomes must remain PENDING EXECUTION.

Test IDAreaTest caseExpected resultActual / status
T01BuildTypeScript and ESLint checkNo errorsPENDING TESTING
T02AuthenticationRegister, login, logout, role redirectCorrect access and error statesPENDING TESTING
T03DatabaseCreate/list/update domain recordsRecords persist and normalisePENDING TESTING
T04Image workflowCapture or select imagePreview and validation workPENDING TESTING
T05AIRun demo and configured API pathCorrect source/status labelPENDING TESTING
T06PrivacyRun utilities and budget trackingExpected mathematical outputsPENDING TESTING
T07Responsive UIReview narrow phone and desktop layoutsNo blocking overflowPENDING TESTING
T08SecurityAttempt unauthorised route and invalid inputAccess denied / validation shownPENDING TESTING
Implementation

Implementation & Documentation

Module IDModulePurpose / evidenceImplementation status
M01AuthenticationEmail lookup, registration, local session, role and approval gating.PARTIALLY IMPLEMENTED
M02User ManagementUser listing, activation/deactivation, and account settings.IMPLEMENTED
M03Patient ManagementPatient dashboard, screening records, cases, and emergency snapshot UI.PARTIALLY IMPLEMENTED
M04Medical Image CaptureBrowser camera capture and image preview; upload/inference pipeline incomplete.PARTIALLY IMPLEMENTED
M05AI InferenceDemo payload plus optional external inference API contract.PARTIALLY IMPLEMENTED
M06ResNet-50 ModelModel metadata and demo labels; weights/training code absent.DOCUMENTED
M07Differential PrivacyNoise, RDP-style accounting, PII stripping, and budget utilities.PARTIALLY IMPLEMENTED
M08Federated LearningNode, round, quorum, privacy, and model data structures.PARTIALLY IMPLEMENTED
M09FedAvg AggregationAggregation method stored as metadata; coordinator absent.PROPOSED
M10Secure CommunicationTLS/gRPC fields represented; transport not present.PENDING VERIFICATION
M11Audit LoggingTyped audit CRUD and client event logging.IMPLEMENTED
M12ReportingCase/report screens and patient-facing summaries.PARTIALLY IMPLEMENTED
M13Control CentreAdmin pages for nodes, rounds, models, privacy, users and audits.PARTIALLY IMPLEMENTED

Technology and platform

TechnologyCategoryPurposeStatus
React + TypeScriptFrontendApplication UI and type-safe modulesIMPLEMENTED
Vite + TanStack RouterPlatformBuild and file-based routingIMPLEMENTED
Tailwind CSSUI stylingResponsive design tokens and utilitiesIMPLEMENTED
Blink SDKData platformTyped database access and project clientIMPLEMENTED
ResNet-50 / PyTorch / OpacusAI / MLNamed model design and documentationPENDING VERIFICATION
FedAvg / DP-SGDFederated privacyDomain terminology and proposed architecturePARTIALLY IMPLEMENTED
gRPC / mTLS / X.509Transport securitySchema/documentation fieldsPENDING VERIFICATION

Security separation

Implemented or visible: route-level role redirects, approval gating, consent fields, audit event writes, PII stripping helpers, app-lock UI, legal pages, and offline state handling.

Recommended / pending verification: password hashing and verification, secure server sessions, server-enforced authorisation, row-level security, actual DICOM sanitisation, mTLS, private CA, certificate revocation, encryption-at-rest, backup/recovery, and threat-model testing.

Deployment planning

Changeover to the System

MediSecure AI is a research prototype. The repository does not establish that it replaces a live hospital information system, so a legacy migration claim would be inaccurate.

Direct changeover
Fast but high risk; not recommended for this prototype.
Parallel changeover
Run beside an existing workflow for comparison.
Pilot changeover
Use a controlled research group before wider use.
Phased changeover
Preparation → pilot → training → acceptance → monitored rollout.

Recommended strategy: preparation, pilot, training, acceptance testing, phased deployment, monitoring, and rollback capability. Any future clinical use requires independent validation, governance, and applicable regulatory review.

Closing record

Project Summary

MediSecure AI currently provides a structured research-oriented web interface and data model for a privacy-aware medical AI concept. It presents roles, cases, screenings, model metadata, privacy metrics, audit events, and offline/PWA support.

The academic boundary is explicit: the repository supports the interface and several client-side utilities, while model training, federated execution, secure node transport, server-side authentication enforcement, and formal experimental evaluation remain incomplete or unverified.

Research and medical safety notice

MediSecure AI is a research and decision-support prototype. It does not replace qualified medical professionals. AI outputs must be clinically reviewed before decisions are made. No clinical certification or regulatory approval is claimed.

Developed by
Anyoha Elvis Chidumebi — Computer Science Student
Supervisor
Dr. Ugo Nwankwo