Table 2: Trial results from using a training set of 21 blank plates and randomly chosen set of 39 patient isolate sample plates and blind-test set of remaining 39 samples from a total of 78 Klebsiella pneumoniae patient isolate plates.

From: High-throughput and automated diagnosis of antimicrobial resistance using a cost-effective cellphone-based micro-plate reader

(A) System Performance
Well Accuracy98.25%
Well Sensitivity98.27%
Well Specificity98.23%
MIC Accuracy95.32%
Drug Susceptibility Accuracy99.55%
Very Major Error Percentage0%
Major Error Percentage0.21%
Minor Error Percentage0.30%
(B) Drug Susceptibility Misdiagnoses
Plate IDDrugAutomated System DecisionDiagnostician DecisionError Type
1AMResistantIndeterminateMinor
5A/SResistantIndeterminateMinor
8MPMResistantSusceptibleMajor
  1. There are 95 applicable wells per plate, with 1 well used as a positive control, 1 well as a positive dye, and the remaining 93 wells for drug testing, yielding a total of 39 × 95 = 3705 applicable wells assessed for turbidity. Each plate tests 17 different drugs and drug combinations for a total of 663 MIC and drug susceptibility decisions. As highlighted in our System Performance summary (A), well turbidity, MIC and drug susceptibility decisions are all comparable to the average performance of our system reported in Table 1. The three specific drug susceptibility misdiagnoses reported in (B) (i.e., plates 1, 5 and 8) are all low-risk misdiagnoses, with 1 major error and 2 minor errors but no very major errors. None of these are for the same drug or plate.