Conventional microscopy is the gold standard for malaria diagnosis. The CellaVision DM96 is a digital hematology analyzer that utilizes neural networks to locate, digitize, and preclassify leukocytes and characterize red blood cell morphology. This study compared the detection rates ofPlasmodiumandBabesiaspecies on peripheral blood smears utilizing the CellaVision DM96 with the rates for a routine red blood cell morphology scan. A total of 281 slides were analyzed, consisting of 130 slides positive forPlasmodiumorBabesiaspecies and 151 negative controls. Slides were blinded, randomized, and analyzed by CellaVision and microscopy for red cell morphology scans. The technologists were blinded to prior identification results. The parasite detection rate was 73% (95/130) for CellaVision and 81% (105/130) for microscopy for positive samples. The interobserver agreement between CellaVision and microscopy was fair, as Cohen's kappa coefficient equaled 0.36. Pathologist review of CellaVision images identified an additional 15 slides with parasites, bringing the total number of detectable positive slides to 110 of 130 (85%).Plasmodium ovalehad the lowest rate of detection at 56% (5 of 9);Plasmodium malariaeandBabesiaspp. had the highest rate of detection at 100% (3/3 and 6/6, respectively). The detection rate by CellaVision was 100% (23/23) when the parasitemia was ≥2.5%. The detection rate for <0.1% parasitemia was 63% (15/24). Technologists appropriately classified all negative specimens. The percentage of positive specimens detectable by CellaVision (73%) approaches results for microscopy on routine scan of peripheral blood smears for red blood cell morphology.
Red blood cell classification in lensless single random phase encoding using convolutional neural networks
