Improving CNN Training using Disentanglement for Liver Lesion Classification in CT

Background: Training data is the key component in designing algorithms for medical image analysis and in many cases it is the main bottleneck in achieving good results.
Recent progress in image generation has enabled the training of neural network based solutions using synthetic data.
A key factor in the generation of new samples is controlling the important appearance features and potentially being able to generate a new sample of a specific class with different variants.

Methods: In this work we suggest the synthesis of new data by mixing the class specified and unspecified representation of different factors in the training data. Our proposed method includes two major components. The first one is the disentanglement process that separates specified from unspecified factors. The second one is the robust training process that makes use of the disentanglement to synthesize new examples with corresponding categorical vectors.

Results: Our experiments on liver lesion classification in CT show an average improvement of 7.4% in accuracy over the baseline training scheme.

Conclusion: In this study we showed that using disentanglement of factors we can synthesize new mixture of samples from different classes that with the corresponding classification target vector can challenge the training process and achieve superior accuracy for liver lesion classification.