Fig. 2.

Analysis pipeline of RT-LAMP-DETR. Image features are extracted from an image of reaction tubes that has been taken by a smartphone camera using the pretrained ResNet50 model, and then used as inputs to the proposed encoder-decoder model which outputs the estimated 1) reaction tubes' locations with their corresponding categories and 2) row and column position indices of the tubes. Each tube's location is represented by a bounding box with its color indicating the predicted category. The number above each bounding box represents the model's confidence in the estimated category (1.0 = highest, and 0.0 = lowest). The row and column position indices of the tubes are used to identify which RNA sample (e.g. from #1 to #16) each tube belongs to. Once the sample is identified, we combine the two estimated reaction tubes' categories with the same identifier to form the final predicted result: COVID-19 positive, COVID-19 negative, RNA absent, and False positive (inconclusive result due to a combination of positive RT-LAMP-XO and negative RT-LAMP-LG). If a sample cannot be identified, its predicted result is marked as Void. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)