Fig. 4: Developing a machine learning clustering-based workflow for automated image deconvolution.
From: A biological camera that captures and stores images directly into DNA
A An automated workflow incorporating outlier detection and reassignment method, machine learning unsupervised clustering technique, full ON-OFF criteria assessment, and cluster grouping for automated image deconvolution. B An example implemented using GMM to demonstrate the results acquired at different phases. The dots in the graphs represent the raw data or the curated data from the 96 wells. The ‘1’ and ‘−1’ obtained from the Local Outlier Factor denote the inliers and outliers detected respectively from the raw data. The curated data were acquired after reassigning the outliers to the nearest values of inliers. The M0–M2 indicate the computed mean values of the individual clusters for full ON–OFF criteria assessment. The deconvoluted image was then plotted based on the three clusters and after clusters grouping into the final binary state image (‘0’: light blue; ‘1’: dark blue) with 3 error bits (orange). C Validation of the automated workflow on other patterns including full ‘ON’ and ‘OFF’ datasets.
