Dung Beetle Optimized LSTM Classifier for the Diagnosis of White Blood Cells

This paper proposes a diagnosis of WBCs using
Dung Beetle-Optimized (DBO) Long Short-Term Memory
(LSTM) classifier. An image is first pre-processed using an
Iterative Mean Filter (IMF) to minimize noise in the image. The
best threshold value to distinguish the foreground from
background is automatically determined using the Otsu's
Thresholding (OTUS) method of picture thresholding. After
that, the feature extraction method of Histogram of Oriented
Gradients (HOG) is employed to extract essential shape and
texture information from images of blood samples. LSTM is
utilized to classify different types of white blood cells with the
highest accuracy or detect abnormalities. The DBO algorithm is
utilized to fine-tune the learning rates, the number of hidden
units, and other relevant parameters of the LSTM. The
simulation is implemented in Python Software, and it achieved
the highest accuracy of 99%, precision of 99%, and recall of
98% in comparison to traditional optimization techniques.