Please enable JavaScript.
Coggle requires JavaScript to display documents.
Peripheral Blood Smear, Note: Overfitting solutions, Mundhra, D.,…
Peripheral Blood Smear
Analysis of PBS
Manual review is considered the gold-standard
Error-prone
Labor intensive
Process of counting different cell types and subtypes
Morphological abnormalities
Blood Cell types
White Blood Cell
platelets
Red Blood Cell
Automated procedure
Increase accuracy
Reduce cost
Faster
Shonit
Hardware component
Low cost microscope
Automatic
Software component
Localize blood cells
Differential count of subtypes
Advantages over traditional
Low cost
Can perform manual and automatic analysis
Captures multiple magnified images
Deep Learning Techniques
Localization
Method for extracting cells
WBC and platelet extraction
Thresholding
Cannot handle variation
Often fixed
Using U-Net Architecture
Used for segmentation mask
Can localize through convolutions
RBC extraction
Image processing
Otsu's Thresholding
Classification
Use of different extractors
In small patches
64x64 RBC
48x48 platelets
128x128 for WBC
Use ensembles of CNN
Challenges
Annotation from medical experts
Variation of subtype classification
Imbalance of samples
Handling overfitting
Shallower architecture
Data augmentation
L2 regularization
Large dropout
Stop when diverging
Results and Discussion
High scores of specificity and sensitivity
Applicable for small laboratories without advanced equipment
Validated with unknown data
Reached 99.5% sensitivity on WBC
Note: Overfitting solutions
Mundhra, D., Cheluvaraju, B., Rampure, J.,
and Rai Dastidar, T. Analyzing microscopic images
of peripheral blood smear using deep learning.
pp. 178–185.