Introduction Cervical cancer mortality in Poland is high. Access to diagnosticians is still insufficient. To deal with this problem a multilayer LBC sample scanner and software was built and implemented, which improved accuracy of diagnostics and limited time of obtaining results. Due to shortage of diagnosticians, a support system based on artificial intelligence algorithms was launched, offering the possibility of remote viewing of scans samples and medical history of the patient. The final diagnosis is always made by cyto-screeners on the basis of system results and cyto-screeners analysis.

Methods The software is based on the artificial neural network (U-NET architecture) designed to recognize suspicious regions and a neural network (VGG) allowing to determine the type of disorder. A machine learning element (fuzzy K-Means) was added - responsible for the fusion of the patient‘s medical history with the neural network system results. A differentiating algorithm included is the crucial part of the system to increase sensitivity of the method especially in recognizing HSIL, ASC-US, ASC-H, Ca Plano.

Results 3161 (LBC) samples were evaluated by cyto-sreeners. Cytological abnormalities were found in 458 (14.3%) cases. Selected samples with diagnosed abnormality were a model to teach the artificial intelligence algorithms. Preliminary results obtained so far indicate 94–97% compliance with results obtained using standard methods. Implementing additional differentiating algorithm has improved results to the level of 96–98% compliance.

Conclusion/Implications Further refinement of neural networks is necessary to improve sensitivity and specificity. A study with a larger sample size will be conducted to evaluate the software.