I am working on a project to classify lung CT images into benign and malignancy using a 3d CNN, for making my work a novel work I want to combine or compare my CNN classifier with another classifier, but I don't have any idea which to choose. could please give me your suggestion about that or any other idea that make my work novel?
You can use your CNN classifiers to compare with other CNN classifiers on some benchmark medical datasets.
CNNs are the best solution to image classification, thus you could compare CNNs with different architectures or different parameters (as Shaopeng Liu said): number of layers, slide length and size of the filter, number of the features...
Therefore you can use architecture with pre-trained convolutional layers for comparison.
You can compare with LSTMs, Logistic Regression, GRUs.
Or you can do transfer learning, i.e
CNN (learn features) +SVM (do classification)
thank you all of you, I am master student, and want to make a contribution in my research not only comparing CNN with itself. I am very thankful if you have any idea sharing it with me.
thank you, this is new for me maybe it helps in my case. If you any article about this type of network, please share the link.
In this paper ( aparticular application of the concept), they learn a road network with a local focus, as opposed to parralel pixel classification :
http://news.mit.edu/2018/new-way-to-automatically-build-road-maps-with-aerial-images-0417
there are other use in reccurent networks applied to spatial data.
To add a novel aspect to your project, consider integrating your 3D CNN classifier with a graph neural network (GNN) for lung CT image classification. GNNs have shown promising results in various domains, and by combining their ability to capture spatial relationships and patterns with the deep learning capabilities of CNNs, you can create a hybrid model that leverages the strengths of both architectures. The GNN can be used to model the connectivity between different regions of interest within the lung, capturing subtle dependencies that may contribute to the presence or absence of malignancy. This integration will not only enhance the performance of your classifier but also introduce a novel approach by exploiting the complementary nature of CNNs and GNNs, leading to potentially improved accuracy and interpretability in lung CT image classification.
- Badges
- Science topic