PL EN
Contact
Contact

CASE STUDIES

Classic Analytical Methods Over AI – Precise Segmentation of Eye Structures

Implementation:

2023 – 2025

Industry:

MedTech / Ophthalmology / Diagnostic Imaging (Digital Health)

Client's Business Profile:

Manufacturer of advanced diagnostic equipment for ophthalmology and provider of Enterprise-class medical software for clinics, hospitals, and research institutes specializing in advanced vision therapies.

Implementation:
Industry:
Client's Business Profile:

2023 – 2025

MedTech / Ophthalmology / Diagnostic Imaging (Digital Health)

Manufacturer of advanced diagnostic equipment for ophthalmology and provider of Enterprise-class medical software for clinics, hospitals, and research institutes specializing in advanced vision therapies.

??
??
??

Challenges

Modern ophthalmology relies on Optical Coherence Tomography (OCT) technology, which allows for non-invasive imaging of retinal cross-sections. However, when interpreting this data, the client encountered significant medical, scientific, and technological barriers:

  • Low precision and time-consuming diagnostics: Manual analysis of microscopic eye structures such as retinal layers or blood vessels was prone to errors and extremely time-consuming for ophthalmologists, delaying the diagnostic process.
  • Bottleneck in drug-gene therapies: Monitoring these therapies required absolutely repeatable, micron-level precision over time. Traditional methods did not allow for error-free, quantitative tracking of changes in the patient's eye structure before and after drug administration.
  • Modern eye diseases and niche therapies suffer from a lack of sufficient unique OCT scans. This prevented proper training of Deep Learning models.
  • The 'black box' problem of AI in medicine: Artificial intelligence algorithms are non-deterministic – the same model can yield different results with minimal changes in image noise, which, in medical device certification (MDR) and critical medical decisions, caused resistance and a lack of trust from medical personnel.

Solution

  • Subjecting raw OCT scans to multi-stage digital filtration (including advanced anisotropic filters, speckle noise reduction, and histogram equalization). This process maximizes contrast between individual eye layers without distorting their actual physical boundaries.
  • Development of a system utilizing mathematical methods, including differential geometry, for automatic segmentation of blood vessels and delineation of retinal layer boundaries in 2D and 3D images. The system enables precise localization of analyzed structures, potentially with single-pixel accuracy.
  • The absence of AI models eliminates the need for costly hardware solutions. The entire system was encapsulated in an efficient API module, allowing for immediate integration with the client's existing medical software.

What the Client Gained

Physicians gained a tool with the highest analytical precision

This enabled reliable and early detection of retinal diseases and precise management of patients in drug programs.

Reduced analysis time for complex spatial eye structures

This increased the throughput of ophthalmology clinics.

Independence from AI limitations

The client mitigated the problem of insufficient training data and received a solution that operates with the same, perfect effectiveness.

Predictability, repeatability, and auditability

This facilitates passing rigorous medical certification procedures and guarantees full operational transparency.