Automatic determination of a biological condition of a subject from ferning patterns

A method comprising:

• air-drying a fluid sample from a subject in a microfluidic device to provide a dried fluid sample having an associated ferning pattern;
• imaging the dried fluid sample at a camera to provide a sample image containing the ferning pattern;
• providing the sample image to a computer vision model;
• and
• determining, at the computer vision model, if the ferning pattern is indicative of a biological condition of the subject, wherein determine if the ferning pattern is indicative of the biological condition at the computer vision model comprises providing the sample image to a convolutional neural network that applies, at each of a plurality of convolutional layers, a depth-wise convolution kernel to each of a plurality of channels within the network and a point-wise convolution that applies a one-by-one kernel across all of the plurality of channels within the network.

A system comprising:

• an optical assembly comprising at least one lens;
• a microfluidic chip comprising a channel to hold a thin layer of a saliva sample from a subject and configured to engage with the optical assembly such that the reservoir is aligned with the optical axis of the optical assembly, the channel being configured such that the thin layer of the fluid sample forms a ferning pattern via air drying;
• and
• a mobile device, comprising:
• a camera aligned along the optical axis of the optical assembly;
• a processor;
• and
• a non-transitory computer readable medium storing executable instructions for determining if the subject is ovulating from images captured at the camera, the executable instructions comprising:
• a camera interface configured to instruct the camera to capture the images;
• a convolutional neural network that determines if the ferning pattern is representative of ovulation by the subject in the at least one image, the convolutional neural network comprising:
• a plurality of convolutional layers, each applying a depth-wise convolution kernel to each of a plurality of channels within the network and a point-wise convolution that applies a one-by-one kernel across all of the plurality of channels within the network;
• a fully connected layer that is connected to a last of the plurality of convolutional layers;
• and
• a classification layer that is connected to the fully connected layer;
• and
• a user interface that displays the output of the convolutional neural network on a display of the mobile device.