Dataset for quantification of CFUs using AI

In recent years, smartphones and software have developed to help to
automate imaging processes and thereby support digitalization in clinical
microbiology laboratories. Due to the time-consuming nature of manual
bacterial colony counting, diverse apps for smartphones have been
developed to quantify the number of colony-forming units (CFUs) on agar
plate images.  Currently, at least four different apps with a
colony counter function exist. Our study aimed to compare the accuracy of
the apps compared to human visual manual counting. The images were
acquired using four different colony counters with three apps on an iPhone
11 Pro Max (CFU.Ai, Promega Colony Counter, APD Colony Counter App PRO)
and one app on a Samsung Galaxy A52 (@BactLAB). Two different measurement
conditions were investigated. Firstly, standardized measurement in a
handcrafted wooden apparatus, and secondly “freestyle measurement”
manually holding the plate. The apparatus measurements were performed
against a black background, apart from @BactLAB on a white background (as
per the manufacturer’s instructions). We used three different types of
media: blood agar, chrome agar, and LB. Four Escherichia
coli isolates were used. We prepared a bacterial suspension of
each strain, made a six-fold serial 1:10 dilution in saline and spread the
dilutions “1:104”, “1:105”, and “1:106” at three
different solution volumes (100μl, 50μl, and 25μl) homogeneously
across the agar plates. We included a total of 108 agar plates. This
dataset contains all analyzed images. It may possibly be useful in order
to assess and improve colony counter apps in the future. On the one hand,
the dataset contains raw data (images before processing), on the other
hand, it contains screenshots of our app measurement and thus enables
accountability of our results (screenshots of all apps were uploaded apart
from @BactLAB as from its screenshots nothing can be derived).