A Regression Neural Network Platform for Real-Time Liver Graft Analysis

Summary

INNERLUXES built a European medical-device manufacturer an AWS-based, HIPAA/GDPR-compliant platform that uses a regression neural network to estimate liver steatosis from spectrometry scan data in real time. With an offline-capable iOS app for procurement surgeons, web review tools, Power BI dashboards, and an admin model-training panel, the solution lets surgeons assess extracted livers objectively without invasive biopsies.

About the Client

The Client is a medical device manufacturer headquartered in Europe that has designed and delivered organ-preservation medical technology for more than twenty years.

The Need: A Secure, Data-Driven Platform for Liver Transplant Decisions

The Client launched an R&D initiative to support clinical decision-making during liver procurement. Developing a new spectrometry-based product, it wanted to complement the hardware with a digital solution that could quantify steatosis levels quickly and objectively, letting surgeons assess extracted livers without invasive methods such as biopsies or frozen-section histology. The Client engaged INNERLUXES to design a predictive software component for the spectrometer — a model that estimates steatosis percentage from scan data — usable across hospital settings, with surgeons performing and reviewing scans via a mobile app and administrators and analysts managing the platform through web interfaces.

To succeed, the system had to run reliably in high-stakes, time-sensitive environments with varying internet availability, support geographically distributed teams, capture data consistently across hospital sites, and ensure traceability for audits and clinical validation — while establishing a foundation for continuous improvement of the prediction models through centralized data collection and training.

An AWS-Based Predictive Platform for Liver Graft Assessment

The project began with a discovery phase led by INNERLUXES's business analyst and solution architect, who captured detailed functional and non-functional requirements for every role in the liver-assessment workflow — surgeons of different seniority, hospital-level administrators, and global platform managers. To align on UX and guide development, the team produced storyboards, user-flow diagrams, and functional maps for each module.

Based on the findings, INNERLUXES proposed a modular, AWS-based architecture meeting the project's scalability, availability, and data-privacy requirements, including HIPAA and GDPR compliance. At its core, a neural-network regression model predicts steatosis; INNERLUXES used Apache PredictionIO for model orchestration, Spark for distributed processing, HBase for event storage, and HDFS for training datasets and models, with model metadata, performance metrics, and evaluation results in Elasticsearch for monitoring and iteration. All data exchange was encrypted over secure HTTPS and JSON APIs, and real-time workflow notifications were delivered via push to keep medical staff promptly updated. The platform's predictive capabilities are delivered through three tightly integrated components:

• An iOS mobile app — the primary data-collection tool during procurement. Junior surgeons connect to a spectrometer, capture liver-tissue scans, and submit results for senior review; the assigned senior surgeon is notified, evaluates the scan, and makes the final procurement decision. The app works offline and syncs when back online.
• A web application — for reviewing scan data, making procurement decisions, and managing surgical teams. Junior and senior surgeons view team-relevant scans, filter data, export records, and manage team membership. Surgeons and hospital administrators each get role-specific Power BI dashboards — surgeons see scan outcomes, donor information, and team activity; administrators monitor platform-wide usage, team composition, and performance.
• An administrative panel — for global admins to manage platform settings, assign roles across hospitals, and oversee usage, with a graphical interface to manage the predictive model (launching training cycles, evaluating new datasets) and reporting dashboards for scan volumes, prediction accuracy, and admin activity across all facilities.

INNERLUXES delivered the platform in phases, starting with an MVP covering the core mobile workflows, role management, initial Power BI dashboards, and the first iteration of the predictive model. Later releases added user-feedback collection, GPS-based scan tracking, and integration with a national healthcare registry.

Better Collaboration, Decision-Making, and Traceability

The Client received a cloud-based platform that supports real-time liver assessment during organ procurement — enabling structured data capture, remote collaboration among surgical team members, and objective decision-making without invasive procedures. Its modular AWS architecture ensures scalability, integration flexibility, and secure, HIPAA/GDPR-compliant data handling, positioning the Client to evolve its product ecosystem and explore the digital transplant-tools market by complementing its medical hardware with intelligent decision-support software.

Technologies and Tools

AWS (Amazon S3, Amazon CloudFront, Elastic Load Balancer, EC2, EC2 Security Groups), Apache PredictionIO, Apache Spark, Apache HBase, HDFS, Elasticsearch, MongoDB, PostgreSQL, Microsoft Power BI, Swift.