1 INTRODUCTION

Trichomoniasis (or trich for short) is the most

prevalent non-viral sexually transmitted infection

2021a). It is caused by infection with a protozoan

0.4–0.9) in men (Rowley et. al. 2019). Because no

recommendations are available for general screening

for TV, the epidemiology of trich has largely come

from population-based and clinic-based surveillance

studies (CDC 2021).

the process generally follows a benign course and is

genital symptoms, and untreated infections might

infection can cause adverse pregnancy complica-

(sensitivity 38–68%), culture of specimens (44–

non-viable organisms in the specimen. The FDA-

diagnostics, but they use large robotic platforms and

are time-consuming, thus not feasible for nPOC.

Moreover, the result is typically yes or no, rather

than being quantitative. The histochemical methods

that binds the trich specific molecules enables an

challenging task to automatically detect the rare TV

microorganisms such as white blood cells, epithelial

multiband image suffers from degradation due to

sample variability, and preparation variability.

microorganisms including the bacteria, epithelial

cells, TV, white blood cells, and yeast), and blue for

channel 4 (TV). The pseudo-color for display is

emission waveband, which is further elaborated in

section 2.1. The non-specific staining can be seen in

a faint eccentric, elongated nucleus. The organism

varies in size but is typically around 10 μm in length

and 7 μm in width, slightly larger than the white

study. The platform runs with dual resolutions, four

analyze vaginal microbiome swab samples for

strong cytokeratin and yeast areas are masked out

using a set union:

labelled. Each blob will be further quantified by the

then imaged with a 40X objective and examined to

2.2.2 TV Segmentation with Multiscale

In order to accommodate for the shadow depth of

focus with the 40X objective imaging, which may

thresholding recovers the cells with undefined edges

objective functional, i.e., maximizing the Dice shape

similarity between the multiscale edge synthesized

segmentation and automatic thresholding

segmentation of an input image 𝐼 , expressed as

follows,

where 𝐼

of the threshold value 𝑇 to an integer set {1, 2, …,

L}. 𝐻∙ is the Heaviside function acting as a

thresholding operator. 𝐶 denotes the edge

synthesized contours by morphological linking of

operator; 𝑆

segmentation is given by the fusion 𝐻

may contain false contours resulted from the

clustered edges of noise. Moreover, the algorithm

visualization of the targets of interest, including

background subtraction. The TV spectral indices are

using a radial basis function (RBF) kernel.

illustrated in Figure 6.

well. The sample is fixed with methanol for 5

adhesive antigen antibody is applied. The sample is

incubated for 15 minutes at room temperature. Then

The overall time for slide preparation is less than 30

(a) (b)

(c) (d)

Figure 5. (a) Composite image of the pseudo-colored four-band IF micrograph at a 40X objective. (b) Segmentation

contours of the TV and its nucleus, with the ROI box overlaid on the composite image. (c) Spectral index TVSI

4 CONCLUSIONS

We have developed a set of novel spectral indices

for a quick, robust search and quantitative

of specimen samples for assisting trich diagnosis in

a point-of-care setting. We demonstrate that the

incorporates a cost minimization with an implicit

shape regularization and self-validation. It is capable

levels and edge strengths. Moreover, this method is

promising results and achieved a high sensitivity and

eliminates user variability in slide reading and offers

the ability to archive images. The system enables a