Graph Regularized Hierarchical Diffusion Process With Relevance Feedback for Medical Image Retrieval

Abstract

Befitting from the interpretability and the capacity in capturing the underlying manifold structure, diffusion process (DP) has attracted increasing attention in the field of image retrieval. Within it, hierarchical diffusion process (HDP) has achieved satisfactory results in retrieved performance and complexity. However, the existing hierarchical diffusion process methods only diffuse the affinity values in low-level visual space without considering the high-level semantic information, which cause the problem of semantic gap. To overcome these problems, we propose a Graph Regularized Hierarchical Diffusion Process (GRHDP) method with relevance feedback, and apply it to retrieve medical images. The proposed algorithm firstly establishes a hierarchical structure of the images in medical image database and spreads the affinity values among query images and top-layer images by graph regularization diffusion. Then relevance feedback is introduced to adjust the similarity between query images and retrieved images in top layer, and the affinity values are diffused again according to labeled information of feedback. Finally, the similarity between queries and others in database can be obtained by interpolating the diffused results on the top layer from top to bottom. The experimental results show that our proposed GRHDP with relevance feedback has achieved better retrieval performance than manifold ranking and regularized diffusion process (RDP) when returning top retrieved images.

Design of a High Efficiency Cross-Domain Medical Image Retrieval System Using Augmented Transfer Learning

“…Results of retrieval are used for retuning the model, and estimating the best hyperparameters for efficient model performance. Due to which, the proposed model is 18% more accurate than GRH [2], 36% more accurate than RaLM [5], and 34% more accurate than DHV [16], while it has 14% more recall than GRH [2], 19% more recall than RaLM [5], and 29% more recall than DHV [16] for different types of medical images. Similar observations were obtained for precision, fMeasure and AUC performance, which indicates that the model is useful for a wide variety of medical image retrieval applications.…”

“…This improvement in precision is due to use of application specific retrieval method selection, which makes the model highly useful for clinical applications. From the fMeasure evaluation, it is observed that the proposed model has 14% more fMeasure than GRH [2], 18% more fMeasure than RaLM [5], and 26% more fMeasure than DHV [16] for different types of medical images. This improvement in fMeasure is due to improvement in precision & recall performance.…”

“…A wide variety of system models are available for medical image retrieval, and each of them have their own performance enhancements with respect to their application of use. For instance, work in [2,3,4] proposes Graph Regularized Hierarchical Diffusion Process with Relevance Feedback (GRH DP RF), convolutional neural network with supervised hashing, and densely-connected Multiple Magnification hashing for retrieval of histopathological images. These models have good performance for a specific pathological image type, and thus cannot be extended to other applications without compromising on retrieval quality.…”

“…and the proposed model for comparison. Based on this configuration, table1showcases the values of accuracy (A) for different test set sizes (TSS), and different algorithms, Accuracy Evaluation for Different Algorithms on Different Test Set SizesFrom the accuracy evaluation, it is observed that the proposed model is 18% more accurate than GRH[2], 36% more accurate than RaLM[5], and 34% more accurate than DHV[16] for different types of medical images. This improvement in accuracy, makes the model highly useful for clinical applications.…”

“…Average area under the curve (AUC) values for different algorithmsFrom the AUC evaluation, it is observed that the proposed model has 18% more AUC than GRH[2], 36% more AUC than RaLM[5], and 33% more AUC than DHV[16] for different types of medical images. This improvement in AUC is due to use of application specific retrieval method selection, and its continuous tuning, which makes the model highly useful for clinical applications.…”