This study by Fanizzi et. al. was an attempt to use machine learning with images of colorectal liver metases (CLRM) from the Cancer Imaging Archive, to assess whether disease progression could be predicted within three years of initial diagnosis, especially when considering the differences of CLRM presentation in men and women. 197 images of the liver, tumor, and hepatic portal vein in patients with CLRM were first stored in an ITK MetaImage format, then converted to DICOM Segmentation Objects (DSOs) and pre-processed by altering contrast and brightness values of specific pixel values that indicated areas such as soft bone or tissue. Using the Python library PyRadiomics, researchers were able to extract data from the images regarding volume, surface area, spatial relationships, and texture. An analytical framework was then developed and applied first to the entire selection of images as a whole, and then to male-only and female-only subsets; the goal for the framework was to identify features that were significant in at least 40% of images and eliminate any confounding cases with the help of five different machine learning algorithms. The data would help train an ensemble machine learning model that would use area under the curve (AUC) values multiplied by the probabilities of specific outcomes based on each feature present in the image, in both the combined-gender dataset and the individual gender-specific datasets. Results showed that in a sample of 117 men and 79 women, 70 men and 48 women showed disease progression in three years, though statistical tests such as the Wilcoxon Mann-Whitney test and the Chi-Square test showed no statistically significant difference in predictive power of the algorithm based on gender differences. Of the clinically features that were identified by the framework, 38 signficant features were identified in the overall dataset, 16 were identified in the male subset (where 11 of those 16 overlapped with the overall dataset's features), and 60 were identifed in the female subset (where 10 of those 60 overlapped with the overall dataset's features). Approximately 44 confounders in the entire dataset, 27 confounders in the male dataset, and 19 confounders in the female dataset were found. The model trained on the mixed-gender dataset had an accuracy of 69.8%, very similar to model trained on the male-only dataset with an accuracy of 73.7%; the model trained on the female-only dataset had an accuracy of 80.5%. The lack of overlap of features between men's and women's respective datasets indicates a significant difference in CLRM presentation across genders, most likely due to increased immune response in the female liver and differing effects of estrogen vs testosterone.