*3.2. Classification Speed*

All models were able to classify the full test set in under 1 min. To import all the necessary Python libraries, build the architecture, load the architecture with the pretrained fertility classification weights, and ge<sup>t</sup> the model's fertility prediction for the 157 images in the test set took a mean time (over five repetitions) of 28 s for the bespoke architecture, 41.7 s using the VGG-16 architecture, 38.5 s for ResNet-50, and 36.5 s for InceptionV3. This compares with an estimated 5 min 14 s taken for one human expert to classify the same number of images (figure determined by multiplying 157 by the mean time of 2 s taken for four trained technicians to classify 30 random ovary images).
