9. Conclusion - djkurran/Automated-framework-for-evaluating-microwave-and-multi-modality-breast-images GitHub Wiki


An automated image analysis workflow was presented that extracts measurable information from regions, objects, and responses within images. This information was used for quantitative analysis and for qualitative interpretation of images using visualization tools. It has been reported in a recent literature review that there are significant discrepancies in how the findings of studies are evaluated and reported [29]. In addition, there is no standardized set of metrics for evaluating images. The workflow that was presented addresses this deficiency by standardizing the processing and analysis of images acquired from multiple imaging techniques.

Specifically, the segmentation methodology presented in [17] that featured an unsupervised machine learning approach to partition images reconstructed with microwave tomography into regions dominated by tissue types was generalized. The generalized form of the technique is capable of processing images reconstructed over a broad range of modalities including radar, ultrasound, and multimodality approaches. The workflow was further adapted to operate on volumes and 3D masks, representing responses, and to accommodate the ability to compare and quantify differences between two reconstructions. The generalization of the workflow was extended to analysis operations by broadening the metrics capabilities. This enabled the assessment of reconstructions containing dielectric properties, backscattered energy intensities, and information related to tissue density.

The robustness and flexibility of the workflow is illustrated through application to reconstructions with a wide variety of characteristics. Its effectiveness was demonstrated with multiple examples that focused on quantifying changes to images due to enhancements of the reconstruction algorithm or perturbations of a parameter used by the reconstruction operator.