Summary
Information about the tools used by HEAP
The goal of HEAP is to help professionals find the target eggs quickly. The internal implementation method is to provide tools to predict the target from the pictures uploaded by the user, so that the user can quickly identify the target eggs from the candidate eggs.
The tools provided by the platform mainly use deep learning to predict the position of candidate eggs from the map.At present, two models of SSD300 and U-net are used. Before predicting candidate eggs from the graph, we need to prepare a training data set for the model to learn identification rules.
Here is a summary of the number of parasite egg types provided by the platform, the number of tools provided, and the number of models.
Table 1. Introduction table of identification tools provided by the platform.
| Number of parasite egg types | Number of tools | Number of models |
| 17 | 3 | 22 |
|
|
|
Here is an introduction to the types of parasite eggs identified by the platform, the source of the training data set, and the training data set information.
Identified parasite egg type
This platform currently provides identification of 17 kinds of parasite eggs, and the available tools are shown in the table. Based on the scalability of the platform, more identifiable types of parasite eggs and available tools will be added in the future to help users find the target eggs.
Table 2. Identification of parasite eggs provided by the platform.
| Parasite egg category | SSD300 | U-net | Faster R-CNN |
| Trichuris trichiura egg | ✔ | ✔ | ✔ |
| Ascaris lumbricoides egg(fertilized) | ✔ | ✔ | constructing |
| Diphyllobothrium latum egg | ✔ | ✔ | ✔ |
| Enterobius vermicularis egg | ✔ | constructing | ✔ |
| Taenia/Echinococcus egg | ✔ | ✔ | ✔ |
| Fasciola hepatica egg | ✔ | ✔ | ✔ |
| Schistosoma japonicum egg | ✔ | constructing | ✔ |
| Toxocara canis egg | constructing | constructing | ✔ |
| Clonorchis sinensis egg | constructing | constructing | ✔ |
| Fasciolopsis buski egg | ✔ | constructing | constructing |
| Paragonimus westermani egg | ✔ | constructing | constructing |
| Echinostoma spp. egg | ✔ | constructing | constructing |
| Hymenolepis diminuta egg | ✔ | constructing | constructing |
| Hookworm egg | ✔ | constructing | constructing |
| Hymenolepis nana egg | ✔ | constructing | constructing |
| Schistosoma haematobium egg | ✔ | constructing | constructing |
| Schistosoma mansoni egg | ✔ | constructing | constructing |
Training data set source
The source of the data set used for training is the slide provided by the Parasitology Department of Chang Gung University. Scan the slides through the instrument to obtain images of different depth of fields. The table shows the parameters used to scan the slides.
Table 3. Equipment and parameters for scanning slides
| Parameter | Value |
| Scanner Agent: | Yuanli Instrument |
| Microscope equipment: | Olympus BX53 |
| Scanning software: | cellSens Dimension |
| Objective magnification: | 10X |
| Eyepiece magnification: | 10X |
| Exposure time: | Automatic |
Training data set information
The images scanned from the slides need to be manually labeled to generate a data set for training the model. During model training, feature values are extracted from pictures of different sizes according to the model settings. The training data set information corresponding to the identification tools trained with different models are shown in tables. Tables on the website indicated how many images captured and the total depth of field covered in each specimen. This is an important feature of HEAP to simulate the actual experience of microscopic observation. (Note: Training data can be obtained from the "Download Pre-train Data" page.)
SSD300
Use pictures with a size of 300*300 and manually labeled .xml files with labelImg for training.
Table 4. Data set information for various types of worm eggs using SSD300 training
| Parasite egg category | Slide name | depth of fields(μm) | Number of images | Number of training sets |
| Trichuris trichiura egg | original | 110 | 12 | 340 |
| Ascaris lumbricoides egg(fertilized) | (Csp-1) | 80 | 9 | 1434 |
| Diphyllobothrium latum egg | (92W5257) | 80 | 9 | 159 |
| Enterobius vermicularis egg | (0822) | 80 | 8 | 638 |
| Echinococcus granulosus egg | (PS1710) | 100 | 11 | 290 |
| Fasciola hepatica egg | (30-6406) | 100 | 11 | 346 |
| Schistosoma japonicum egg | (PS1301) | 100 | 11 | 159 |
| Fasciolopsis buski egg | Buski_egg | 100 | 11 | 112 |
| Paragonimus westermani egg | (PS1415) | 330 | 34 | 100 |
| Echinostoma spp. egg | (E-16) | 100 | 11 | 88 |
| Hymenolepis diminuta egg | (92W5341) | 100 | 11 | 285 |
| Hookworm egg | (HL E-3) | 90 | 10 | 80 |
| Hymenolepis nana egg | (92W5361) | 100 | 11 | 84 |
| Schistosoma haematobium egg | (92W5123) | 100 | 11 | 100 |
| Schistosoma mansoni egg | (92W5153) | 90 | 10 | 80 |
U-net
Use 512*512 pictures and manually labeled .json files with labelme for training.
Table 5. Data set information for various types of worm eggs using U-net training
| Parasite egg category | Slide name | depth of fields(μm) | Number of images | Number of training sets |
| Trichuris trichiura egg | original | 110 | 12 | 485 |
| Ascaris lumbricoides egg(fertilized) | (Csp-1) | 80 | 9 | 655 |
| Diphyllobothrium latum egg | (92W5257) | 80 | 9 | 117 |
| Echinococcus granulosus egg | (PS1710) | 100 | 11 | 265 |
| Fasciola hepatica egg | (30-6406) | 100 | 11 | 287 |
Faster R-CNN
Use 300*300 pictures and manually labeled .xml files with labelImg for training.
Table 6. Data set information for various types of worm eggs using Faster R-CNN training
| Parasite egg category | Slide name | depth of fields(μm) | Number of images | Faster R-CNN (default) Number of training sets | Faster R-CNN (clear) Number of training sets |
| Trichuris trichiura egg | original | 110 | 12 | 619 | 332 |
| Diphyllobothrium latum egg | (92W5257) | 80 | 9 | 58 | 25 |
| Enterobius vermicularis egg | (0822) | 80 | 8 | 480 | 268 |
| Echinococcus granulosus egg | (PS1710) | 100 | 11 | 177 | 45 |
| Fasciola hepatica egg | (30-6406) | 100 | 11 | 217 | 121 |
| Schistosoma japonicum egg | (PS1301) | 100 | 11 | 97 | 60 |
| Toxocara canis egg | (92W5823) | 100 | 11 | 56 | 17 |
| Clonorchis sinensis egg | (PS1218) | 80 | 9 | 674 | 164 |