blood smear malarial parasite detection - stanford universityxj933th8548/... · malarial rbc?...

1
Blood Smear Malarial Parasite Detection Austin Zheng EE368, Department of Electrical Engineering, Stanford University Motivation Parasite Detection and Isolation Pipeline Further Work Results (5 test images) HSV conversion H, S and V components (histogram equalization applied) Thresholding Candidate analysis Nuclei detection Attempt cell segmentation of the region. Malarial RBC? Spurious signal? Free-floating parasite? Partial to complete automation of blood smear counting of red blood cells (RBCs) infected with malaria parasite (P. falciparum) Valid RBC Spurious stain Better circle detection better handle elliptical cells, poor segmentation, blended shapes, etc. Automatic preprocessing parameter calculation resilience to noise, differing lighting, poor image quality, etc. Challenges: Low image quality (lighting, focus) Flaws in smear procedure Overlapping cells Clutter Non-circular (ovoid or deformed RBCs) Original image Cell edge detection Better nucleus discrimination Schizont Trophozoite Ring Note: Edge nuclei intentionally omitted. Schizont form (large dark mass) currently produces poor results Segmentation may be impossible in some cases. Other heuristics may be necessary in conjunction with cell segmentation. Image 1 3 successfully detected clusters, 6 spurious/missed clusters Image 2 4 successfully detected clusters, 0 spurious/missed clusters Image 3 6 successfully detected clusters, 0 spurious/missed clusters Image 4 2 successfully detected clusters, 12 spurious/missed clusters Image 5 5 successfully detected clusters, 6 spurious/missed clusters Post-processing can remove redundant cell signatures from well- formed output.

Upload: others

Post on 22-Aug-2020

16 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Blood Smear Malarial Parasite Detection - Stanford Universityxj933th8548/... · Malarial RBC? Spurious signal? Free-floating parasite? Partial to complete automation of blood smear

Blood Smear Malarial Parasite Detection Austin Zheng

EE368, Department of Electrical Engineering, Stanford University

Motivation Parasite Detection and Isolation Pipeline

Further Work Results (5 test images)

HSV conversion

H, S and V components (histogram equalization applied)

Thresholding

Candidate analysis Nuclei detection

Attempt cell segmentation of the region.

Malarial RBC? Spurious signal?

Free-floating parasite?

Partial to complete automation of blood smear counting of red blood cells (RBCs) infected with malaria parasite (P. falciparum)

Valid RBC

Spurious stain

Better circle detection better handle elliptical cells, poor segmentation, blended shapes, etc.

Automatic preprocessing parameter calculation resilience to noise, differing lighting, poor image quality, etc.

Challenges: • Low image quality (lighting, focus) • Flaws in smear procedure • Overlapping cells • Clutter • Non-circular (ovoid or deformed RBCs)

Original image

Cell edge detection

Better nucleus discrimination

Schizont Trophozoite Ring Note: Edge nuclei intentionally omitted.

Schizont form (large dark mass) currently produces poor results

Segmentation may be impossible in some cases. Other heuristics may be necessary in conjunction

with cell segmentation.

Image 1 3 successfully detected

clusters, 6 spurious/missed

clusters

Image 2 4 successfully detected

clusters, 0 spurious/missed

clusters

Image 3 6 successfully detected

clusters, 0 spurious/missed

clusters

Image 4 2 successfully detected

clusters, 12 spurious/missed

clusters

Image 5 5 successfully detected

clusters, 6 spurious/missed

clusters

Post-processing can remove redundant cell signatures from well-

formed output.