Agricultural technology has evolved significantly since the 1800s. However, with respect to monitoring and farming of aquaculture such as oysters, industry is still stuck in the past. This project aims to modernize oyster monitoring and identification with machine learning. We developed a program capable of accurately identifying and classifying oysters in three states. They include closed, meaning they are healthy, or open for long periods of time, indicating mortality, and indeterminate. We also increased the scope of existing software with a website for real-time identification that is accessible to the public. In the process, we added support for video annotation, which was previously difficult to perform for those with no prior coding experience.1

Bacteria Based Cancer Therapy (BBCT) is a unique form of cancer treatment where bacteria is genetically engineered to induce production of the anticancer drug from within the tumor without affecting normal tissues. This project aims to study and develop image processing and machine learning algorithms to segment and track the size of the tumor regions affected by the anti-cancer bacteria based on the diffusion-weighted MRI images (DWI) collected from animal tumor models over the course of treatment, providing methods of evaluating its effectiveness. We employ emerging image processing and machine learning techniques that are implemented in MATLAB and Python to create algorithms consisting of pre-processing, processing, and post-processing steps, with the goal of calculating the destructed tumor tissue volume at each stage of the treatment. To automate the pre-processing and identification steps of the tumor region, a convolutional neural network called a U-Net is utilized to predict a binary mask to identify and crop the raw MRI images around the tumor prior to the processing and the post-processing steps, then the algorithm calculates the total volume of the tumor, dead tissue, and live tissue. The U-Net is trained using binary masks of the tumor regions alongside the original MRI images. The developed algorithms allow researchers to quantitatively determine the effectiveness of the BBCT by tracking the volume of dead tumor tissues. Ultimately, this research could lead to an automated 3D volumetric tumor tissue destruction measured in vivo longitudinally in the DWI experiments.