The kingdom of bacteria is a very diverse group of organisms characterized by high phenotypic variability. This feature is often used in clinical diagnosis. The spirochaetes are a microbes with a characteristic spiral shape of a flagella located within the periplasmic space. Nowadays, there is a high demand for creating a rapid and sensitive method for their detection as many of them performs a high pathogenic risk. Currently used methods lays on combination of clinical examination results, serologic and cultivation methods. There can be also used Polymerase Chain Reaction (PCR) method if needed. Unfortunately this combination can be very time consuming and require a lot of money. This research presents a novel, semantic segmentation neural network architecture designed to quickly create a classification mask, outputting information about the position, shape, and possible afiliation of detected elements. The evaluation method isbased on a light microscope imagery and was created to overcome above mentioned problems. Used abstract classes contains erythrocytes, spirochaete and background. The resulted mask can be later mapped to a human-readable form with the inclusion of colors, next to an original image. Such approach allows for semi-automatic recognition of unwanted objects, however still giving the final verdict to the specialist. Developed solution has achieved a high recognition accuracy, while the computer power requirements are kept at a minimum. The proposed solution can help reduce misclassification rates by providing additional data for the doctor and speed up the entire process with the early diagnosis made by a neural network.
IVUS 2022: 27th International Conference on Information Technology *Corresponding author. † These authors contributed equally. $ michal_wieczorek@hotmail.com (M. Wieczorek); natjia@wp.pl (N. Wojtas)
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License in our paper we propose a quick and simple method of CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g ACttEribUutRion W4.0oInrtekrnsahtioonpal (PCCroBYce4.0e).dings (CEUR-WS.org) evaluating the presence of spirochaete bacteria, that may 2 after each block. The internal signal addition within the convolutional blocks contains several branches with diferent layers count and combinations of batch normalization layers. The final block output is combined using concatenate and add layers for better signal fidelity. Sample scheme is presented in Fig. 4. The training has been optimized using NAdam algorithm with learning rate of 0.0078 and the selected loss function was Categorical Cross-entropy with custom class weights computed before training to balance the training.
To improve model’s performance in terms of final accuracy performance and the training times the NAdam training algorithm has been used. The formula can be described as follows: = 1−1 + (1 − 1),
2 = 2−1 + (1 − 2) , where parameters are constant hyper-parameters and is the current gradient value of an error function. Values and are used later for computing the correlations marked as ˆ and ˆ according to below equations: ˆ = (1 − 1) + 1+1 ˆ =
1 − 2 .
ˆ = −1 − √2 +
Finally, using previously calculated variables, the final formula can be defined as: speed up the diagnosis, give the doctors a valuable clue and save money and health of the patient. This can be especially valuable for busy, first contact clinics as a primary method of evaluating the presence of spirochaete microorganisms, whose presence can be predicted after a basic physical examination, before some more detailed and expensive tests.
The concept of microbe detection can be approached using different various techniques, however some of the most accurate methods include visual classification.
Normally such process is performed by a human specialist and contains manual checking of hundreds of objects within previously selected frames per patient.
This process is extremely slow and requires full focus of the doctor for the whole time in order to avoid misclassification and oversight.
As in the nature of such examination are high repeatability and small amount of additional stimulus, it is very difficult for any human being to maintain the peak detection performance during the whole process, especially considering some external factors like tiredness, small amount of time, very little visual differences between microbes or lack of special interest in the field.
Above mentioned problems lead to high average detection error rate.
More common deep learning techniques contain rectangle masking, such as in [6] and [7], however for this task the output needs to be more precise. Because of that as a partial solution to this issue in this research a custom semantic segmentation neural network architecture has been created. It provides additional data, in the form of a mask with initial elements classification, to the doctor next to the original image for easy and fast verification. Such approach can highly reduce error rate by providing additional diagnosis and pointing out suspicious elements, as well as speed-up the entire diagnosis process by reducing the time needed to analyse the image.
Additionally, by choosing segmentation architecture over the classical rectangle masking one, the classification is made per-pixel and thus there is clarity and accuracy improvement on images with higher amount of overlapping objects.
During the research phase the most suitable one, containing masked images of the spirochaete mi
Final architecture is based on the U-Net shape and the croorganisms mixed with the red blood cells was the
final parameters were selected empirically. Final shape “Bacteria detection with darkfield microscopy” dataset
is presented in Fig. 3. The input layer has a shape of gathered and annotated as part of a bachelor thesis of
256x256 and is reduced by max-pooling by a factor of university Heilbronn, Germany. The dataset contains
(
There were several factors needed to be taken into
consideration while searching for the dataset:
• The data have to contain microscopy imagery of
both microbes and healthy cells,
• The dataset has to be free for academical use,
• The images need to have appropriate masks.
Algorithm 1 NAdam training process
1: Generate random weights,
2: while global error value < _ do
3: Shufle the training dataset,
4: for each batch inside training dataset do
5: Compute gradient vector g on the batch,
6: Update vector eq. (
With that in mind some compromises has been made, mainly on the training length side, however after the final reduction the model consists of 7,921,534 parameters and weights around 94MB. The evaluation times are below 0.1 second on the GPU and around 0.87 second on the CPU per image.
The training plots are presented in Fig. 1.
Although the data are high quality and each image consists of many microbes and red blood cells, the number of training examples is relatively small to train a highly accurate model without the use of data augmentation.
After several trials including variety of simple image transforms, as well as state of the art methods based on Generative Adversarial Networks (GAN), such as Principal Component Resampling presented in [8], the best combination in this case includes horizontal and vertical flip, random image rotation and random zooming.
The network originally outputs the data in a form of a 366 images from the darkfield microscopy with manually two dimensional matrix with sparse representation of created masks labelling 3 abstract classes: background, classes using integer values. Such data are optimal for spirochaete and erythrocytes. being stored and analyzed by the computer, however presents no useful value for the non-technical user and requires further processing to create an informative 3.1. Data augmentation image. That’s why, in order to make it readable, the matrix has been expanded by 3 additional color channels and integer values from the [0, 2] range has been mapped to red, green and blue channels. Based on basic human psychology blue has been chosen as a background, green as harmless blood cells and red as dangerous microbes. Such prepared mask is presented next to the original image for fast and easy validation by the user.
Other methods of visualization has been considered, such as merging both original and mask into one image, however the level of clarity has been highly reduced and the validation became much more dificult as some of the original data has been compromised.
During this research all computations were made on a PC with specification below: • CPU: Ryzen Threadripper 2950X 16c/32t, • RAM: 128GB, • GPU: NVidia RTX 3090 24GB.
During this research one of the main goals was to create not only an accurate model but also to reduce its memory and power requirements to the bare minimum. Such approach is crucial, as it helps to spread the use of similar models in real-world applications. Although computer hardware is becoming more powerful each year, very little people, especially in smaller clinics are Sample final results produced by the network can be seen in Fig. 2.
In the above examples there can be seen that, although the network in some cases struggles to find t he exact shape of the microbe or the red blood cell, it is still able to perform really well in most cases, even the more extreme ones where the image is not the highest quality, there is high amount of overlapping elements, the contrast is very low or the microbe is small relative to the whole image.
This paper presents a novel solution for fast spirochaete detection using a custom Semantic Segmentation Neural Network. The output is presented in a clear and easy to understand way, also allowing for quick validation if (a) Accuracy Plot (b) Loss Plot needed. Although the training has been performed on a powerful GPU, the evaluation could be also done on a CPU from the budget level computer, making it accessible for almost everyone. Presented solution is able to speed up the detection time by providing additional data to the original image, helping the human performing the evaluation spot potential bacteria. This could not only allow for testing more animals at the same time but also drastically reduce the costs of such operation.
In the future there are many paths of improvements both in terms of functionality and accuracy performance. One of them is expanding the current dataset with new images captured on more diverse conditions, such as more noisy backgrounds, diferent bacteria shapes, lower contrast ratio between elements, etc. This approach would lead to much higher accuracy on validation data as the network will understand the wider context, thus the feature extraction should work correctly on more cases than the current model. Another way of improving the network would be by not only adding more images to the training set but also by expanding the number of abstract classes fit the potential new dataset some changes in the Deep providing examples of other microbes. This would lead to Learning architecture could be necessary to further better understanding of the world by the model but improve the accuracy. could also reduce the misclassification rate. To better