Non Hodgkin's Lymphoma Classification using Improved Predator Optimization Based Densenet121 Model

Abstract

CAD system (computer aided diagnosis) assists medical experts in NHL (Non Hodgkin's lymphoma) diagnosis for making better decisions. In the case of NHL diagnosis, the manual analysis required more time. A few issues with the current methods included low accuracy, increased computational complexity, low dependability, increased feature dimensionality, and increased time consumption due to inadequate hyperparameter optimization. Hence, this work presents a CAD approach that ensures efficient and accurate diagnosis of NHL at the beginning stage. The proposed work has the different stages like pre-processing, optimal feature extraction and classification. The input images are resized and the features from the images are extracted and classified. This process is carried out by the DL (deep learning) model DenseNet121 with IPO (improved predator optimization) approach. The hyperparameters like batch size, neurons in the dense layer and learning rate are optimized by the IPO which minimizes the over-fitting and complexity. The analysis is demonstrated on the malignant lymphoma classification dataset and achieved a better accuracy of 98.6%.