Using Computer Vision for Lung Lesion Detection

Cancer remains a widespread and deadly disease. A common diagnos7c approach for early  detec7on is the use of CT scans for lesion detec7on. There is poten7al for computer vision  methods to serve as a diagnos7c tool to assist radiologists in evalua7ng CT scans. This project uses data from the NIH Cancer Imaging Archive to develop a computer vision model capable of identifying lesions in human lung CT scans. The final model achieved recall comparable to, or in some cases better, than human radiologists, while maintaining a high F1 value. Its classification threshold can also be tuned on the use case, enabling the model to reach up to 95% recall. Future work should incorporate additional training data and explore applications across other issue types. Overall, this model is effecive and shows potential to be used as a diagnostic tool in clinical setting.