• Home
  • Malaria
  • Other diseases
  • Publications
  • About
OCTOPI
  • Home
  • Malaria
  • Other diseases
  • Publications
  • About

MALARIA

Need for accurate and scalable diagnostic tools

MANUAL MICROSCOPY
  • WHO gold standard
  • Low sensitivity and low limit of detection 
  • Error-prone, workload dependent
RAPID DIAGNOSTIC TEST (RDT)
  • Simple and easy to use
  • Not quantitative
  • False negatives (due to gene deletions)
MOLECULAR TECHNIQUES (e.g. PCR)
  • Most sensitive
  • Expensive, resource-intensive
  • Widespread implementation challenging
AUTOMATED MICROSCOPY - OCTOPI 
  • ​Visual identification and quantification of fluorescently labeled parasites
  • Consistent performance; operator independent
  • Integration of expert review and large scale annotation 
  • Patient-level sensitivity and specificity > 97% at parasitemia cut-off of 15 parasites/µL*
  • Estimated limit of detection ~ 12 parasites/µL*
  • Scanning and processing > 500,000 RBCs/minute
*Based on clinical data from recent study

Processing pipeline

  • Consistent blood smearing with Inkwell, where each slide contains a monolayer of up to 13-16 million red blood cells (RBCs)
  • Simple, quick and convenient staining using DAPI (fluorescent stain)
  • High throughput scanning (> 500k RBCs/minute) and machine learning based segmentation and image classification
Picture
Procedure for applying Octopi to malaria diagnosis.

Real-time malaria detection 

  • Automated blood smear scanning and machine learning based red blood cell and malaria parasite quantification
  • Identification and reporting of fluorescently labeled parasites (along with “parasite prediction score”) for convenient review and validation
  • Image acquisition, cell segmentation and parasite classification completing in ~ 1 second per field of view (containing up to 8-10k RBCs)

High throughput imaging and analysis

  • Monolayer of red blood cells and DAPI stained parasites covering most of the smeared area (unlike traditional thin smears with useful information in only the feathered edge)
  • Analysis of millions of red blood cells in a single slide, i.e. orders of magnitude larger than traditional techniques, thus significantly improving the limit of detection for malaria microscopy  
  • Fluorescence microscopy (along with differential phase contrast imaging) enables larger field of view using 20× microscope objective, instead of 100× oil objective 

Machine learning cellular-level performance

  • ResNet-18 model with one round of retraining, resulting in false positives < 5 parasites/μL (average)
  • Per-parasite level false negative rate < 8% in our test dataset
  • Estimated limit of detection LoD ~ 12 parasites/μL for P. falciparum malaria parasites
Picture
​

Patient-level performance

  • Independent dataset with 73 patient/donors and 190 blood smears (parasitemia for malaria positive patients ranging from 16 parasites/µL to around 130,000 parasites/µL)​
  • Sensitivity = 100% and specificity = 97.4% at parasite cut-off of 15 parasites/µL
  • Sensitivity = 97.1% and specificity = 100% at parasite cut-off of 20 parasites/µL
Picture
Check out our latest pre-print

Want to collaborate or use Octopi?

Connect with us
Picture
Cephla Inc.
705 N Shoreline Blvd, 
Mountain View, CA 94043
USA
Contact us

Home

Malaria

Other diseases

Publications

About

Copyright © 2025
  • Home
  • Malaria
  • Other diseases
  • Publications
  • About