Automated Lymph Nodes Classification using Factorized Convolution-based CNN
Hitesh Tekchandani, Shrish Verma, Narendra Digambar Londhe · 2021 Fourth International Conference on Electrical, Computer and Communication Technologies (ICECCT) · 2021
Lymph nodes (LNs) are an essential component of immune system which filter out harmful substances. Any disease of LNs leads to serious medical complications and may reduce fighting spirit of human body. Quantitative assessment of LNs like short axis measurement, plays an important role in evaluation of oncological therapy responses. Selection and analysis of pathological LNs in Computed tomography (CT) images is complex task as it requires the vigilant study of CT images. Hence, in this paper, an automated method based on Convolutional neural network (CNN) has been designed to classify LNs in pathological and non-pathological groups. Large kernel sizes in CNN increases trainable parameters and hence makes it more probable for overfitting. Hence, authors have designed and implemented factorized convolution-based CNN model on machine augmented LNs dataset for identifying pathology. Four variants of convolutional filters based on various level of factorization, are designed and applied for classification of abdominal LNs. The best achieved accuracy is 96.38%.