Deep learning for medical imaging

Building an automated system for assisting radiologists in detecting various pathologies in MRI scans.
Deep learning hero image

Magnetic resonance imaging (MRI) is one of the most advanced noninvasive medical diagnostic methods which produces images of cross-sections of organs in three main planes and their combinations, enabling for more accurate assesment and diagnostic of various diseases.

Deep learning for medical imaging, part 1

Detailed description of machine learning methods and techniques used in our newest medical imaging project.


MR is often a method of first choice for diagnostic deseases of central nervous system organs, as well as various diseases and states of the locomotor system.

While the hazards of other imaging methods that use ionizing radiation (X-rays, CT) are now well-understood and controlled, MRI may still be a better choice in many different scenarios. On the other hand, it is a relatively expensive and time-consuming technique. It generates and captures huge amounts of data containing extremely valuable information, at a pace far surpassing what traditional manual methods of analysis can process. The analysis of MR scans require highly specialized experts that manually review and mark regions affected by some sort of pathology.

Algorithms and processes that could at least partially automate this work could bring be very valuable in clinical practice.


The goal of this project is to produce a Software-as-a-Service (SaaS) system for automating segmenting and classification of orthopedics MRI, focusing on knee scans, using Machine Learning (ML) techniques. Algorithmic methods for MRI analysis fall into two general categories: classification and segmentation.

Deep learning for medical imaging, part 2

The role of data preprocessing and segmentation for improved knee pathology classification in magnetic resonance imaging.

Classification assigns a label to an MRI series — normal/abnormal, level of severity, or a diagnosis. Segmentation is the process of delineating the boundaries, or “contours”, of various tissues and processes. Our system aims to contain functionality of both methods, and serves as a prototype of a „virtual assistant“ in interpreting MR scans to help radiologists in their everyday work.


Finalized system includes functionality for acquiring imaging results via DICOM standard, data pre-processing, processing of images using appropriate ML models, and displaying results in an user-friendly and intuitive fashion. We have applied Class Activation Map (CAM) overlays on top of original scans to show which areas contributed the most to the classification outcome, signalizing which anatomic structures should be evaluated further for problems

Deep learning for medical imaging, part 3

The role of data preprocessing and segmentation for improved knee pathology classification in magnetic resonance imaging.

We are planning to continue development of this system by extending its scope to include support for processing scans of other body regions, as well as support for unlimited number of users using cloud-computing technologies.

Our tech stack

Book a free consultation

Let us know what would you like to do. We will probably have some ideas on how to do it.