ecDNA are small extrachromosomal DNA created by excess DNA damage of the linear genomes. These particles often replicate proto-oncogenes and thus are shown to result in and increase in drug-resistant cancers. The current method of studying ecDNA and the effects various therapies have on the ecDNA count and quality is by using FISH (Fluorescence In-Situ Hybridization). This results in images which trained humans can discern ecDNA from. However, this process of detection requires strong human discretion and a variety of tools to segregate true ecDNA from noise, often taking up to an hour to identify all the positive locations in a single image. We hope to create a deep learning model which is able to train from the labels made by humans previously to learn a more robust object detection scheme, which can be generalized to all images. This would allow us to study drug effects on ecDNA in a fraction of the time while reducing systematic and random biases in enumerating ecDNA results from these trials.