Case Studies / apoQlar
MEDTECH · MEDICAL IMAGING & SURGICAL PLANNING
Automating manual anatomical segmentation with AI for surgical planning
Before every surgery that relies on MRI or CT imaging, someone has to manually outline the relevant anatomical structures – bones, vessels, brain ventricles – slice by slice. It’s painstaking, time-consuming, and a bottleneck in clinical workflows. We built a series of AI models that automate this segmentation for apoQlar’s VSI HoloMedicine® platform, turning hours of manual work into seconds.
Client: apoQlar GmbH – creator of the VSI HoloMedicine® platform, using extended reality devices for 3D surgical planning and patient education.
KEY RESULTS
Hrs→Sec
Manual segmentation work reduced from hours to seconds
Traceable
Full documentation of datasets, training, and evaluation
MRI+CT
Models for both imaging modalities across multiple structures
Azure ML
Full documentation and reproducibility on Microsoft Azure
INDUSTRY
MedTech
USE CASE
Anatomical segmentation for surgical planning
AI APPROACH
U-Net variants + deep learning
IMAGING
MRI & CT scans
PLATFORM
Azure ML + MS InnerEye
INTEGRATION
VSI HoloMedicine® (XR devices)
The challenge
apoQlar’s VSI HoloMedicine® platform allows surgeons to visualize patient-specific anatomy in 3D through extended reality headsets – using real MRI and CT data projected into the surgeon’s field of view. It’s a powerful tool for surgical planning and patient education. But it depends on one critical input: segmented anatomical structures.
Segmentation – the process of identifying and outlining specific structures like bones, blood vessels, and brain ventricles on medical scans – was being done manually. Radiologists or trained technicians would go through imaging data slice by slice, tracing each structure by hand. For a single patient, this could take hours. It was the single biggest bottleneck between receiving imaging data and having a usable 3D surgical model.
The core problem: every 3D surgical visualization depended on manual segmentation that took hours per patient. This manual bottleneck limited how many cases the platform could handle and slowed down clinical workflows at a point where speed matters – before surgery.
What we built
Working with apoQlar’s team and consulting medical professionals across multiple specializations, we developed a series of AI models that automate the segmentation of anatomical structures from MRI and CT scans.
Multiple anatomical structures. The models don’t handle a single use case – they segment different types of structures: bones, blood vessels, and brain components including ventricles. Each required its own model variant trained on specialty-specific imaging data and validated by the relevant medical professionals.
State-of-the-art neural network architectures. We used variations of the U-Net architecture – the standard for medical image segmentation – adapted and optimized for each anatomical target. The models consistently achieved high segmentation accuracy across diverse patient imaging data.
Comprehensive documentation and reproducibility. In healthcare AI, how models are built matters as much as how they perform. We built the entire pipeline on Microsoft Azure Machine Learning and Microsoft InnerEye, ensuring full documentation of dataset characteristics, training procedures, evaluation metrics, and deployment steps. Every model is fully reproducible – the same inputs produce the same outputs – and the entire development process is traceable end to end.
Direct integration with VSI HoloMedicine®. The segmentation output feeds directly into apoQlar’s extended reality platform, where surgeons can interact with the resulting 3D anatomical models through XR headsets including Meta Quest and other devices. The AI doesn’t just produce data – it produces data that’s immediately usable in the surgical planning workflow.
The results
BEFORE
Manual slice-by-slice segmentation taking hours per patient. Limited throughput. Inconsistent results depending on who performed the annotation. A bottleneck before every surgical case.
AFTER
Automated segmentation completing in seconds. Consistent quality across patients and imaging types. Fully documented with reproducible training pipelines. Directly integrated into 3D surgical planning.
The manual segmentation bottleneck was eliminated. What previously took hours of specialist time now takes seconds – with consistent, reproducible results backed by comprehensive documentation of the entire training and evaluation pipeline. Surgeons get their 3D planning models faster, and the platform can scale to handle significantly more cases.
For enterprise MedTech organizations, this project illustrates a pattern that applies far beyond healthcare: wherever a skilled professional spends hours on a repetitive, structured task – especially one that sits in the critical path of a larger workflow – AI can often reduce that to seconds while improving consistency and creating a full audit trail.
As a related workstream on the same platform, we also integrated ShareMedix into VSI HoloMedicine® to solve another manual bottleneck: anonymizing operating room videos and images before they can be shared with colleagues, used in medical presentations, or published. Doctors wearing masks, glasses, and headlamps in OR footage make standard facial recognition software unreliable – ShareMedix was trained specifically on medical-setting imagery to handle these conditions, replacing what was previously a time-consuming manual anonymization step that discouraged doctors from sharing clinical content at all.
Technology used
More Case Studies
See how we’ve helped other companies
AUTOMOTIVE · LEADING LUXURY MANUFACTURER
Intelligent virtual assistant replacing manual planning queries across SAP and cloud systems
Product planners spent hours manually querying SAP BW and multiple data warehouses for every decision. We built a bilingual voice-and-text assistant that retrieves planning data on demand – no system expertise needed.
Hrs → Sec
Data retrieval
DE + EN
Voice & text
SAP BW
Integrated
MANUFACTURING · RADAWAY
Making email-based order processing reliable with LLMs
Staff were manually reading customer emails, identifying products, and entering orders by hand. We turned a promising AI prototype into a production system that handles it end to end, across languages, formats, and attachments.
-90%
Manual intervention
95%+
Match accuracy
LOGISTICS · FR. MEYER’S SOHN
Eliminating manual data extraction from thousands of daily shipping emails
Operations staff were manually reading German and English logistics emails to pull out routing and scheduling data, every single day. We built an AI pipeline that extracts, structures, and delivers the data automatically.
–80%
Manual effort
2 langs
DE & EN
On-prem
Deployed