Brain metastases are a significant cause of patient morbidity, with almost half of all cancer patients developing the debilitating complication. Magnetic resonance imaging using structural T1 is the clinical standard for detecting their formation. Another type of magnetic resonance imaging sequence called diffusion weighted imaging is often acquired and can be used to display apparent diffusion coefficient maps. These maps provide valuable quantitative imaging data on the microstructure of the brain. Additionally, the use of machine learning for quickly analyzing large amounts of data, coupled with complex image data features in the form of radiomics, has allowed for the extraction of textural information previously hidden within medical images. The research presented in this thesis investigated the use of machine learning and radiomic methods on diffusion imaging to create a model that can facilitate earlier brain metastasis detection when compared the conventional imaging methods. A feasibility study was conducted to quantitatively characterize the unique metastatic signal of brain metastases as they form using longitudinal ADC imaging and first order statistics. Distinct changes in local diffusion were found within clinical metastatic regions and motivated further research analyses. Based on these preliminary results, the methodology was expanded to include higher order radiomic features and machine learning methods. Several classifiers were trained using radiomic values calculated from longitudinal ADC maps, clinical metastatic contours, and healthy contralateral control areas. The ensemble classifiers showed superior accuracy in differentiating healthy brain tissue from metastatic tissue. With the knowledge gained from these first two projects, the radiomic and machine learning pipeline was expanded to scan the entire brain for areas of early metastatic formation. Using a patch-based image sampling technique based on anatomic regions of the brain, and radiomic features filtered using stability analysis, the trained models showed up to 75% accuracy in detecting the formation of brain metastases earlier than structural contrast-enhanced T1. The research presented in this thesis highlight a novel clinical application of diffusion-based magnetic resonance imaging for the early detection of brain metastases. Future work will be focused on improving the detection models and integrating them into clinical software for prospective validation and testing.