*5.2. Colorimetry*

One class of optical biosensors is the colorimetric biosensor. Currently, the applications of machine learning to enhance the performance of bioreceptor-free colorimetric biosensors are limited. This limitation is because the colorimetric biosensors (most notably lateral flow assays) mostly utilize bioreceptors (e.g., antibodies, enzymes, and aptamers) [98]. One example of such a bioreceptor-free biosensor is non-invasive plant disease diagnosis by Li et al. [49]. They utilized an array of plasmonic nanocolorants and chemo-responsive organic dyes that interact with volatile compounds from the plant. Their technique is similar to Enose and Etongue since it is a fingerprinting approach to the array response for classification. They used PCA, but do not cite an actual classifier, although they give performance metrics such as accuracy. At this time, it is unclear how the classification was performed on the PCA-transformed data.

Most colorimetric biosensors do not require machine learning due to their simplicity for readout. However, the arrays of bioreceptor-free (semi-specific) colorimetric sensors require machine learning-based classification in a way similar to Enose and Etongue. In these instances, they will likely benefit from the same treatment, namely dimension reduction by PCA and SVM classification.

**Figure 10.** A quantitative phase image of a label-free specimen is virtually stained by a deep neural network, bypassing the standard histological staining procedure that is used as part of clinical pathology. Reproduced from [59] without modification under Creative Commons Attribution 4.0 License.
