FAST (Focused Assessment with Sonography in Trauma) is an ultrasound procedure performed to examine for free fluids such as blood after trauma. The procedure takes a set of skills to perform to get comprehensive results. To gain accurate and comprehensive results it is important to train and evaluate individual operators. Moreover, it is a very tedious and long process for preceptors to evaluate each individual operators’ skills. The aim of this project is to aggregate all key points of interest or in other words highlights from ultrasound scan videos into one single video clip to help preceptors evaluate the operator’s performance. The approach was to use the MobileNet V2 convolutional neural network architecture for binary image classification of highlight frame and non-highlight frame images. The dataset used consists of FAST ultrasound videos of each vital region (perihepatic, perisplenic, pericardial, pelvic) performed by 6 expert participants, 15 intermediate participants, 14 novice participants and 9 returned novice participants. The results showed increasing performance with added trainable parameters and data augmentation using the MobileNet architecture, but the accuracy was still poor. The area under the curve remained constant at 0.5 throughout all experiments. The achieved performance is not yet sufficient for clinical practice. The results indicate using a bigger dataset sample size or a different convolutional neural network architecture, such as InceptionV3 or ResNet, will yield better results.