AI and Transfer Learning–Based Framework for Efficient Classification and Detection of Lyme Disease
Pramit Brata Chanda, Saikat Das, Sharanya Bhattacharya, Souhardya Biswas, Subir Kumar Sarkar · 2024
Lyme disease has become a vector-borne Skin disease throughout the United States, Europe, and some parts of Asia. This requires proper medical examination of symptoms and other diagnoses to detect the disease and, if left untreated, regions can spread to other major organs of the human body. In this work, the analysis is performed on different images of Lyme skin infections that are caused by specific groups of ticks, to build several transfer learning– and artificial intelligence–based frameworks for predicting if the infection is Lyme or not. Today, artificial intelligence has a crucial role in predicting diseases and reducing human error. The AI-based transfer learning framework is very effective for predicting Lyme disease with good accuracy. Here, a sequential model is created with transfer learning models in Keras and lightweight machine learning models in Scikit Learn which are trained on the dataset of images and assessed the performance and accuracy. As the volume of the medical images collected is quite small, we have experimented with other deep learning technologies such as “transfer learning” with “fine-tuning” models that are trained on the same dataset. To counter possible overfitting, pure machine learning algorithms such as logistic regression, k-nearest neighbors, and Naive Bayes are also applied in this study. Here, the gathered datasets of all the metrics of several models are compared with the predicted scores. So, the VGG, ResNet, MobileNet, and sequential CNN-based AI models are providing more than 80% accuracy rates significant in predicting and classifying diseases effectively.