In a recent study published in the Medicine in Novel Technology and Devices Journal, researchers used a large dataset containing images of skin lesions from monkeypox (mpox) patients to develop a machine for detecting mpox. We developed a learning-based detection tool.
study: Deep learning-based monkeypox virus detection using skin lesion images. Image credit: sulit.photos/Shutterstock.com
Background
Mpox is a zoonotic systemic disease caused by the monkeypox virus (MPV). orthopox virus A genus of the Poxviridae family.
Until early 2022, the disease was endemic in West and Central Africa, but as of late 2022, monkeypox cases have been reported from more than 100 countries outside endemic areas, putting recent MPV in North America and Africa. The spread to Europe is considered a global pandemic.
The disease causes fever, headache, muscle pain, swollen lymph nodes, and rashes and lesions on the palms, soles, face, mucous membranes of the mouth and genitals.
The rash begins on the soles of the feet and palms, spreads to the eyes, genitals, and mouth, progressing from usually flat or patchy to hard, raised lesions called papules that eventually fill with pus and form pustules.
The current standard method of detecting monkeypox is to use the polymerase chain reaction (PCR) test, but results are inconclusive because the virus stays in the body for a short period of time or because it is not accessible in rural or remote areas. You often don’t get it.
However, artificial intelligence and machine learning techniques offer faster and more accessible methods of diagnosing disease.
About research
In this study, we developed a model based on a deep learning method for detecting mpox using skin lesion images taken by a regular smartphone camera. This work aimed to accurately detect mpox using various deep learning techniques such as AlexNet and GoogLeNet.
They also compared the performance metrics of other machine learning models used to diagnose mpox in terms of accuracy, recall, accuracy, and f1 score.
The training dataset consisted of 228 images, of which 102 were images of measles and the remaining 126 were images of measles and chickenpox lesions. It consists of 1,428 images of mpox lesions and 1,764 photographs of other lesions using various enhancement methods such as translation, rotation, shear, reflectance, hue, contrast, brightness, saturation and scaling. The data set that can be used has increased.
The deep neural network was trained using the training image dataset from Deep Network Designer running in MATLAB 2022. Pilot runs were conducted on several neural networks including Places365-GoogleNet, GoogLeNet, AlexNet, SqueezeNet, Vision Transformer, and ResNet-18.
result
They reported that ResNet-18’s results showed the highest accuracy (99.49%) among all tested neural networks.
The researchers found that ResNet-18 works with higher accuracy than Places365-GoogleNet, Squeezenet and GoogLeNet due to its effective and simple architecture, and can learn complex features of detection methods without requiring a large number of inputs. I think. ResNet-18 has fewer convolutional layers and lower computer memory requirements than other neural networks.
Vision Transformer models were used as an alternative to traditional neural network models, but were found to perform poorly compared to neural network models when using similar training and validation hyperparameters.
This performance difference is likely due to the visual transformation model utilizing a large number of parameters and requiring a large training dataset.
Deep learning methods in medicine offer faster and more accurate examination options. Large volumes of patient data can be efficiently filtered without sacrificing accuracy or time.
Moreover, it is resource efficient and does not require heavy machinery or expensive equipment, making it an ideal mpox detection method for different medical settings and clinics in different regions.
Conclusion
In summary, the researchers used a large dataset of mpox lesions and measles and chickenpox lesions to develop a variety of neural mechanisms to detect mpox cases from images taken by easily accessible smartphone cameras. trained the network.
Overall, the findings showed that the neural network model ResNet-18 performed best with 99.49% accuracy.
Additionally, with other techniques such as locally interpretable model-independent explanation (LIME), medical professionals can use this method to detect mpox and make predictions based on the results of neural network models. may be visually interpreted.
