Classification of Overlapping Cells in Microscopic Cervical Images: A Transfer Learning Approach

One of the leading cause of mortality in women, is Cervical cancers. Timely and accurate diagnosis at earlier stage can reduce the mortality rate significantly. Pap smear is the popular screening test used for diagnosis of cervical cancer. Now a days, deep learning and convolution neural network based techniques are more popular due to excellent performance over traditional machine learning techniques. The aim of this study was to investigate the use of Deep learning based transfer learning for classification of overlapping cells in cervical smear. Transfer learning technique with Alexnet framework is implemented to improve the classification accuracy. Multicell cervical cytology Cervix 93 dataset is used which is publicly available. Hyper parameters of the network are fine tuned to make the model suitable for cervical cancer classification problem. To evaluate the performance of the network statistical parameters like Accuracy, Sensitivity, Specificity, Precision, NPV, PPV, F-score and Matthews correlation coefficient (MCC) etc. are calculated. The proposed network reported outstanding performance with 99.86% accuracy and specificity. The overall performance of the proposed framework is promising and can assist the clinicians in diagnosis process.