Figure 1.
Illustration of the HCS system and the malaria image analysis algorithm.
(A) 384-well screening platform to set up experimental assays. (B) Operetta 2.0 imaging system to acquire image data. (C) Central database to store acquired image data. (D) IM platform to load and analyze image data from central database. (E) The malaria image analysis algorithm diagram from input image data to analysis results. The algorithm is implemented as a plug-in of IM platform.
Figure 2.
Total RBC number estimation process.
(A) A partial region of 400×300 pixel2 of WGA-AlexaFlour 488 fluorescence channel image. (B) Separate RBC regions by Otsu’s threshold method. (C) Classify isolated and clustered RBCs by concave corner point detection method based on boundary curvature measurement and principal axes ratio. (D) Count number and calculate average area of isolated RBCs. (E) Estimate number of RBCs in each cluster. (F) Compute number of total RBCs.
Figure 3.
Parasite signal detection process diagram.
(A) DAPI or Mitotracker fluorescence channel image. (B) Detect local maxima points of intensity. (C) Compute and
of local maximum point intensities. (D) Smooth signals by Gaussian filtering. (E) Separate signals by threshold value
. (F) Compute number, location, area, signal strength information.
Figure 4.
Infected RBC segmentation process diagram.
(A) WGA-AlexaFlour488 fluorescence channel image. (B) RBC edge structure map (white structures), corresponding gradient vector filed (red arrows), and the fitting circle (yellow circle). (C) Normalized vector field of the sum of the gradient vectors along a fitting circle (yellow arrows). (D)-(F) Infected RBC segmentation process. (D) Infected RBC with detected parasite signal points (red dots). (E) Two fitting processes started from different signal points. (F) Fitting result. Inner region of fitting circle is segmented as infected RBC region. Note that two different fitting processes give to same result.
Figure 5.
Decision tree of the life cycle stage classification criteria.
The life cycle stage of parasites in an infected RBC is classified into early ring, ring, trophozoite and schizont by the presence of the detected parasite signals from the DAPI/Mitotracker fluorescence channels, and the number, area and distance between the signals. Note that “DAPI parasite signal” and “Mitotracker parasite signal” in the decision tree represent the parasite signals detected from the DAPI and Mitotracker channels respectively.
Figure 6.
Area distribution and number of isolated RBCs in randomly selected nine images.
Each image has enough number of isolated RBCs to calculate average RBC area with distribution close to a normal distribution.
Table 1.
Comparison results of manual and algorithm RBC counting.
Table 2.
Inspection results of parasite signal detection.
Table 3.
Inspection results of infected RBC segmentation.
Figure 7.
Life cycle stage classification process validation result.
Highly synchronized P. falciparum Dd2 cultures were imaged at 8, 30 and 40 hpi and were analyzed by the malaria image analysis algorithm. The dominant parasite stages at 8, 30, 40 hpi were early rings, trophozoites, schizonts respectively.
Figure 8.
Comparison of EC50 for known antimalarial compounds.
EC50 values of the chloroquine, artemisinin, and pyrimethamine against P. falciparum 3D7 parasites using SYBR I, pLDH, and our assays were determined and compared.
Table 4.
Comparison of EC50 for known antimalarial compounds.