Artificial intelligence helps to evaluate skin allergy tests
Polish scientists have SkinLogic-Solution developed that enables more efficient skin allergy testing and more reliable results. The method uses video and thermal imaging cameras and a system that analyzes the images down to the last pixel.
The authors of the described solution are specialists from the Faculty of Electronics and Information Technology of Warsaw University of Technology, the team of Professor Jacek Stępień (Milton Essex company) and the Military Medical Institute.
The clinical tests gave very good results. The system correctly identifies up to 98% of cases, even rare ones Allergies. In addition, it is with SkinLogic possible to detect lesions with a maximum diameter of 0,3 mm.
Image source: Pixabay
Development and operation of SkinLogic
As pointed out in a press release from the Warsaw University of Technology (WUT), from an IT perspective, SkinLogic is a data processing system. The device consists of a tripod and the cameras mentioned at the beginning. During the tests, the patient's hand must be fixed in the stand. The device takes pictures with visible and infrared light at certain times and records what is happening on the skin fragments treated with allergens. Once the digital documentation is available, it's time to edit the PW algorithm to use.
Importantly, with the usual manual method of measuring allergic reactions (blisters), the result is not entirely accurate. However, when using SkinLogic, the measurement is performed by the algorithm. In addition, the system checks both the size of the reaction and other parameters, such as e.g. B. their shape. The image obtained with the far-infrared spectrum is useful for this.
Analysis of digital material
During the analysis, the images are divided into segments corresponding to the location of the incisions on the skin (each segment can be examined separately). By analyzing the data over time, one can see how the segment has changed.
Where does the input data for the artificial intelligence system come from? They used 1500 allergic skin reaction images (records) that doctors collected during clinical trials in 100 patients. This allowed the algorithm to learn to recognize which image represents an allergic reaction and which does not.
What we get from the camera images are 100x100 pixel images. A doctor examining an allergic blister only has the area visible to the naked eye. We examine every pixel on the images. One could say that a standard diagnosis is based on a single value, while the response tested by artificial intelligence is based on millions of values and recognized combinations", explains Professor Robert Nowak, Head of the Department of artificial intelligence. It would be extremely difficult for a human to find these patterns; a trained algorithm does this job quickly and is very accurate. More data means more noise to eliminate, but the algorithm can handle this problem too. Our system was trained using a set of patterns developed by a medical consortium, so it has a high-quality foundation," adds the researcher.
Improved diagnosis and treatment planning
The system is currently being tested as part of the pre-registration. Once used in clinical practice, it can be an invaluable aid. It means faster Diagnosis, provides more precise results and allows easier consultation with other specialists thanks to the digital acquisition of material.
The article "Thermography-based skin allergic reaction recognition by convolutional neural networks" was published in the journal Scientific Reports in mid-February.