Advancing Monkeypox Diagnosis: A Novel Approach Using a Custom Neural Networks

A few red bumps that might seem innocent at first may hold a darker secret-monkeypox, a viral disease that is a member of the chicken pox family. The issue arises since monkeypox is highly contagious and therefore any close contact with the victim can be detrimental. Often the person can get assured after a series of diagnostic tests and examinations such as PCR and viral culture to confirm the presence of Monkeypox but they involve invasive and painful sample collection procedures, like blood draws which could be painful. Moreover, the process involves multiple tedious steps which can devour a lot of valuable time and are relatively expensive, making it less accessible to all the people. This research aims to make the diagnostic process efficient, cost-effective, and accessible, ultimately improving the identification and management of Monkeypox cases in resource-constrained environments. To improve monkeypox diagnosis, this study utilized a dataset of whole-body skin images captured by a smartphone and developed a novel custom neural network consisting of 41 layers where separable convolutional layers techniques as well as fire module were used alongside the traditional convolution layers. Our proposed model outperformed other state-of-the-art methods with an accuracy of 96.88%, a precision of 95.30%, a recall of 99.24%, and an Fl-score of 97.23%, the classification model shows how accurately it can identify monkeypox cases with the help of computer-assisted techniques using smartphone.