Background Only 1.3% of the sub-Saharan African diagnostic laboratories are performing clinical bacteriology. To improve this, diagnostic tools in LRS should be simple, affordable and maintenance-friendly, in contrast with the expensive machinery used in high-income countries, such as mass spectrometer for identification. Lensfree imaging is a label-free identification technique that can be performed directly on colonies growing on agar plates, with low-cost instrumentation.

Methods We report here the very first clinical assay of a wide-field lensfree imaging device, namely 24mm x 36mm, for identification down to species. Considering this large field of view, several hundreds of colonies can be analysed simultaneously. A database of over 250 clinical bacterial isolates was collected at LHUB-ULB, gathering respiratory (20% of isolates), urine/genital tract (20%) and skin/wound (20%) samples, as well as positive blood cultures (40%). Partially coherent light emitting diodes (wavelengths 550nm and 940nm) illuminated microbial cultures growing on Mueller Hinton agar at 36°C. Clinical isolates were labelled through MALDI-TOF mass spectroscopy. To optimize supervised learning, various deep learning models, pre-trained or not, were developed and compared.

Results Over 11.000 colonies were collected in the database that focused on 5 species (Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Staphylococcus aureus, Staphylococcus epidermidis). From these, different deep learning models were trained with at least 1800 samples per species. As a result, the algorithms yielded unambiguous identification of each species, with at least 90% accuracy.

Conclusion This very first database paves the way towards a future imaging device for the diagnosis of bloodstream infections in LRS, within the SIMBLE project. As a second stage, a second database is to be acquired, in Africa, on positive blood cultures.