One of the most prominent types of dementia is Alzheimer’s disease (AD). It's among the main causes of death in elderly individuals in all developed nations. Numerous deep learning (DL) models for image classifications and object identification have been developed. However, DL techniques prepare the network model from scratch, which has some downfalls including demanding a massive amount of labeled training data source, which might be a trouble in the medical world, one in which practitioners annotate the data, being very expensive, and usually requires high computational resources. Transfer learning methods are currently being utilized to address these difficulties. In this research, a transfer learning-based ResNet 50 model that was pre-trained on the ImageNet dataset was adjusted on different hyper parameters using the ADNI dataset to provide the best possible results. There are four distinct optimizers utilized, SGD, Adagrad, rmsProp, and Adam, as well as two different batch sizes. The results show that the optimizers’ rmsProp and Adamax performed well with batch sizes of 16 and 32 when compared to the SGD and Adam optimizers. For batch size 32, the classification accuracy for AD vs. NC using rmsProp is 74.22%. However, using batch size 16 resulted in a 1.01% relative improvement with 75.65% classification accuracy.
CITATION STYLE
Sethi, M., & Ahuja, S. (2023). Hyper Parameters Tuning ResNet-50 for Alzheimer’s Disease Classification on Neuroimaging Data. In Lecture Notes in Computational Vision and Biomechanics (Vol. 37, pp. 287–297). Springer Science and Business Media B.V. https://doi.org/10.1007/978-981-19-0151-5_25
Mendeley helps you to discover research relevant for your work.