Feature Blending Approach for Efficient Categorization of Histopathological Images for Cancer Detection
Anish Anurag, Rik Kamal Kumar Das, Govind Kumar Jha, Sudeep D. Thepade, Neha DSouza, Chandrani Singh · 2021
Advancements in medical imaging has resulted in efficient diagnosis of lethal ailments like cancer by means of histopathological image data. Rich insights about the impact of the disease can be figured out with careful examination of the histopathological images captured using high end cameras. This paper has attempted to investigate the usefulness of pretrained convolutional neural network features (CNN) for automated classification of the histopathological image categories. MobileNetV2 is considered as the pretrained architecture for CNN based feature extraction. The experimentation process has resulted in designing lightweight blended feature vectors using handcrafted techniques which are of significantly smaller dimension compared to bulky CNN features and has disclosed higher classification efficacy compared to CNN features with reduced computational overhead.