Conference Papers
Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/123456789/28506
Browse
Search Results
Item Nuclei Segmentation using EfficientNetV2 and Convolutional Block Attention Module(Institute of Electrical and Electronics Engineers Inc., 2023) Kadaskar, M.; Patil, N.Cell nuclei segmentation is crucial for developing digital pathology and medical research. It is helpful in numerous applications, including illness diagnostics and other medical therapies. Unfortunately, due to the massive number of nuclei, manual analysis of these image slides takes time and effort. Morphometric appearances add to the complexity. Nevertheless, we upgraded the Nested UNet with an EfficientNetV2S backbone and added a CBAM module in the decoder layers to obtain state-of-the-art performance in medical imaging. This improved model is easier to train, and attention blocks help to tune the retrieved features for better performance. We tested our model using the CryoNuSeg dataset to see how well it performed. Our model scores 0.941 on Dice, 0.605 on AJI, and 0.614 on PQ. © 2023 IEEE.Item Nuclei Classification in Histopathology Images Using Fuzzy Ensemble of Convolutional Neural Networks(Institute of Electrical and Electronics Engineers Inc., 2023) Kadaskar, M.; Patil, N.The development of computational pathology and healthcare studies depends on cell nuclei classification. It has several lifesaving uses, particularly in cancer diagnosis. It is helpful in various applications, including disease diagnosis and medicinal therapy. Due to the number of problems, human examination of these image slides is unpleasant and time-consuming. Morphological traits add to the intricacy. We suggested a fuzzy distance ensemble of Convolutional Neural Networks to achieve state-of-the-art nucleus classification performance. This new model is easier to train and makes more accurate predictions using underlying basic classifiers. To examine how well our model worked, we used the PanNuke dataset. Our model's scores are 0.811 for accuracy, 0.811 for F1, 0.815 for precision, and 0.809 for recall. © 2023 IEEE.