What is a computer aided diagnosis algorithm?
Traditional diagnosis involves radiologists to visually interpret medical images for any suspicious signs. However, due to the large amount of medical images, traditional diagnosis is often challenging and can be subject to detection and characterization errors and inter- and intra-reader variability.
To achieve a more accurate diagnosis, different types of computer-aided diagnosis (CAD) approaches have been developed to assist interpretation of the medical images. It assists radiologists in their interpretation of medical images by scanning and evaluating digital images for conspicuous structures. CAD provides the radiologists of a second opinion in the detection and the diagnostic processes, while the radiologist remains the primary reader and decision maker.
The computer algorithm generally consists of several steps that may include image processing, image feature analysis and data classification.
CAD is beginning to be applied progressively in the detection and diagnosis of various types of abnormalities in medical images made by different imaging modalities.
Challenges in CAD validation
Validation refers to the procedures that measure the accuracy of the algorithm to be evaluated. For this purpose, a large data set of images is required, which contains positive, negative, false positive and false negative cases.
Currently, the medical image transfer for validation purposes mainly takes place via three processes:
the shipment of CDs. This method is time consuming, expensive and can be unreliable
Setting up complex VPN tunnels. These tunnels are expensive and difficult to maintain.
Closed source proprietary tools exist to enable DICOM connectivity. However, these are costly to set up and maintenance is managed through recurrent maintenance fees.
Once transferred, the image needs to be stored in such a way that the research team can access the DICOM tags and store analysis results. This is challenging in the sense that there aren’t any tools that are readily available and let the user interact with the study details openly.
Shortly put, none of the current processes are efficient, unless you outsource and/or use costly proprietary tools.
Where Orthanc and Osimis come into play
A simple, powerful and cost efficient way to manage and share medical images was introduced in 2011, by Sébastien Jodogne, member of the Liège University Hospital staff, who created a medical imaging server named Orthanc, initially to respond to the needs of their radiotherapy department. As the research for these developments had been sponsored by government funds, it was then decided to open up the source code.
Orthanc is an open-source DICOM server for medical imaging and also simplifies research about the automated analysis of medical images. Among its many features, it offers centralised storage in a private or public cloud, providing access to images for parties concerned over the internet, from anywhere!
As Orthanc instances can be set up and configured in a breeze, imaging sites can use it as a means to anonymize and transfer studies automatically to a centralized (cloud based) instance Orthanc instance.
From there, the research team can then access the studies through the REST API, enabling, for example, manipulation of DICOM tags or running the CAD algorithm.
Once the algorithm it is tuned, it can be made accessible via portals that Osimis can customise for the client.
Osimis can help the research team in setting up connectivity with imaging sites and can house a drop zone for research projects.
A man for whom 24 hours in the day would just be enough, if he could go without sleep. He has a seemingly endless stream of ideas and prides himself on always being on top of everything going on in his company. Frederic used to work as a boring banking consultant until divine inspiration sentenced him to a life of penal servitude to the medical imaging business. But that was just after he learnt the dirty tricks of full stack coding during an excellent 2-month bootcamp.