Current Status of Computational Intelligence Applications in Dermatological Clinical Practice
Carmen Rodríguez-Cerdeira1, 2, 3, 4, *, José Luís González-Cespón1, Roberto Arenas1, 3, 4, 5
1 Efficiency, Quality, and Costs in Health Services Research Group (EFISALUD), Health Research Institute. SERGAS-UVIGO Vigo, Spain
2 Dermatology Department, Hospital do Meixoeiro and University of Vigo, Vigo, Spain
3 European Women’s Dermatologic and Venereologic Society (EWDVS), Tui, Spain
4 Psychodermatology Task Force of the Ibero-Latin American College of Dermatology (CILAD), Buenos Aires, Argentina
5 Micology Department, Manuel Gea González Hospital, Mexico City, Mexico
The yeast infections are increasingly frequent and the correct diagnosis consists of the identification of the yeast fungus, which in our case we are going to refer to the different species of Candida. The prescription of a broad-spectrum antifungal without taking into account the etiological agent, leads to an increase in the resistance to these treatments.
The objective of this work is to differentiate Candida albicans from other Candida species (Candida spp.) By means of digital images obtained from the optical microscope.
Material and Methods:
It has reviewed about 100 photographs from patients in our consultations.
In this study we will use the microscopic images of the Candida variety to be processed later with the Octave programming language and its image processing package (image-2.8.0).
Results and Discussion:
This system is able to differentiate Candida albicans from the other varieties of Candida such as C. parapsilosis, C. krusei, and C. kefyr with accuracy.
The candida identifier application, which was designed and programmed in Octave, allows identification of candida species by locating certain geometric descriptors, such as the centroid and the surfaces of circular objects within the images. The program was highly effective for the diagnosis of Candida spp. So, we got a sensitivity and specificity above 90% with the images used.
The results that we obtain from the Candida spp. identifier system that opens the way to be able to work with images obtained from the optical microscope.
open-access license: This is an open access article distributed under the terms of the Creative Commons Attribution 4.0 International Public License (CC-BY 4.0), a copy of which is available at: https://creativecommons.org/licenses/by/4.0/legalcode. This license permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
* Address correspondence to this author at the Department of Dermatology, Hospital do Meixoeiro and University of Vigo, Vigo, Spain;
Tel: 0034600536114; E-mail: firstname.lastname@example.org
Current Status of Computational Intelligence Applications in Dermatological Clinical Practice
Intelligent computer systems already provide support to healthcare professionals.
Computational intelligence has been widely researched and applied owing to its ability to cope with large amounts of clinical data and uncertain information [1López-Rubio E, Elizondo DA, Grootveld M, Jerez JM, Luque-Baena RM. Computational intelligence techniques in medicine. Comput Math Methods Med 2015; 2015196976 [http://dx.doi.org/10.1155/2015/196976] [PMID: 25834633] ]. Computational intelligence uses algorithms based on biological data, primarily on neuronal functioning. The three pillars on which computational intelligence is based are neural networks, genetic algorithms, and fuzzy systems [2Haykin S. Neural networks: A comprehensive foundation 1998.]. Neural networks are algorithms employed for function approximation or classification problems. They include supervised, unsupervised, and reinforcement learning [3Hamet P, Tremblay J. Artificial intelligence in medicine. Metabolism 2017; 69S: S36-40. [http://dx.doi.org/10.1016/j.metabol.2017.01.011] [PMID: 28126242] ].
Constant progress is being made on computational intelligence to improve the prevention, diagnosis, and treatment of diseases in general, and mycosis in particular [4Koulouri A, Kuonen F, Gaide O. Artificial intelligence and the skin specialist. Rev Med Suisse 2019; 15(644): 687-91. [PMID: 30916908] ].
The yeast Candida albicans is the most prevalent pathogenic Candida species (spp.), although the increase in immunocompromised patients has been accompanied by an increase in the diversity of pathogenic strains found as etiological agents of fungal infections [5Hedayati MT, Tavakoli M, Zakavi F, et al. In vitro antifungal susceptibility of Candida speciesisolated from diabetic patients Rev Soc Bras Med Trop 2018; 51(4): 542-5. [http://dx.doi.org/10.1590/0037-8682-0332-2017] [PMID: 30133642] , 6Şular FL, Szekely E, Cristea VC, Dobreanu M. Invasive fungal infection in romania: Changing incidence and epidemiology during six years of surveillance in a tertiary hospital. Mycopathologia 2018; 183(6): 967-72. [http://dx.doi.org/10.1007/s11046-018-0293-2] [PMID: 30168077] ].
The most typically used method to identify Candida spp. is the chromogenic culture medium. An analysis of the growth facilitates the determination of the purity of the colonies, and the identification of Candida spp [7Ouanes A, Kouais A, Marouen S, Sahnoun M, Jemli B, Gargouri S. Contribution of the chromogenic medium CHROMagar(®)Candida in mycological diagnosis of yeasts. J Mycol Med 2013; 23(4): 237-41. [http://dx.doi.org/10.1016/j.mycmed.2013.07.058] [PMID: 24161925] ].
The samples used by us are from this culture medium and were later imaged under an optical microscope. After reviewing several images, we have selected C. albicans, C. parapsilosis, C. krusei, and C. kefyr for our project. The images used in the elaboration of the program correspond to Figs. (1-7).
2. MATERIALS AND METHODS
Approximately 100 photographs primarily from the author’s patients were reviewed.
2.2. Description of the Method
Development of programs for the detection of Candida species
As the characteristics of the different species of Candida are known, we applied digital image processing with Octave. We performed an adaptation of the diagnostic methods of the Candida and their subsequent study under a microscope for the classification of the different species Most of these methods have typical concepts, such as the variety of colors, asymmetry in their internal forms, and abrupt and irregular edges, and are used to identify elements of the said fungi under the optical microscope. These are the primary concepts that are analyzed in the programs performed, and are referred to as the “Candida Identifier.”
The objective of this work is to differentiate Candida albicans from other Candida spp. with an accuracy of approximately 70%.
This general objective is divided into more specific objectives as follows:
Establish the spaces of color and form that will be used in this work.
Define the necessary programs, in Octave environment, such that they can recognize the different species of Candida
Develop a rapid, easy, and economical technique for identifying Candida spp. that is easily implementable and manageable even by primary care physicians.
4. RESULTS AND DISCUSSION
(1) The identification algorithm was performed using the following steps:
Load the images
Go from RGB to grayscale
Binarization of the image
Tag pixel regions to establish neighborly relations
Perform geometric measurements to select descriptors
Choice of patterns to differentiate the species of Candida (Candida spp.)
(2) Later, we will perform the stages of the model comprising the following:
(A) Identification and training criteria
(D) Implementation of the algorithm through Octave and program development
(E) Interface design
4.1. Identification and Training Criteria
Subsequently, with part of the program code, we seek to identify the circularity of each element, which is stored in the circularities variable. The results are weighted with an arbitrary threshold to define the number of circular elements of the images, which is stored in the variable roundObjects. These characteristics allow us to differentiate the species studied. For example, C. albicans exhibits circular elements resembling C. kefyr. Meanwhile, C. krusei exhibits more linear elements.
To calculate of the circularity of rounded objects we use the follow program code in Table 1.
Table 1 Determination of the circumscribed circumference from the centroid of the silhouette.
4.2. Test Three Times Using the Images
In the testing stage, it was possible to verify the application from the original image to grayscale and subsequently to its binary form to determine its geometric properties through the Octave “regionprops” function. This function returns the patterns that characterize each type of fungus, thus allowing for the effective identification of the images. The different steps to be taken are: First of all, we have the original image. The second step is to pass it to grayscale image and finally we get the binarized image.
4.3. To Determine Our Model Behavior When we Apply it to Candida sp., we Constructed a Confusion Matrix (Table 2 &3)
Table 2 Confusion matrix with the different Candida spp.
Table 3 Results of confusion matrix with C. albicans, C. krusei, C. kefyr & C. parasilopsis
4.4. Implementation of the Algorithm Through Octave and Development of the “Candydos” Program
The program has implemented the image recognition algorithm, using the leeImage function. In its first line, the image library of Octave is loaded. This library contains the classes and functions that allow for the processing of mushroom images (Table 4)
4.5. Interface from Database of Java Script Libraries
The RGB image (original image) is passed to grayscale and subsequently to its binary form.
Application interface: First step of the Candida recognition process.
Step two of the Candida recognition process and step three, recognition by “candydos” of Candida spp. In this case, the one identified with correction was Candida albicans.
A major difficulty was in obtaining a large sample and images of the Candida spp. used with similar characteristics, because the time of planting and the technique vary according to its shape and size. The system cannot change the size of the images during the entire process [9Colling R, Pitman H, Oien K, et al. Artificial intelligence in digital pathology: A roadmap to routine use in clinical practice. J Pathol 2019; 249(2): 143-50. [http://dx.doi.org/10.1002/path.5310] [PMID: 31144302] ].
Another difficulty that also influences the classification of the image is the image quality [10Aractingi S, Pellacani G. Computational neural network in melanocytic lesions diagnosis: Artificial intelligence to improve diagnosis in dermatology? Eur J Dermatol 2019; 29(S1): 4-7. [http://dx.doi.org/10.1684/ejd.2019.3538] [PMID: 31017580] , 11Wu M, Yan C, Liu H, Liu Q. Automatic classification of ovarian cancer types from cytological images using deep convolutional neural networks. Biosci Rep 2018; 38(3)BSR20180289 [http://dx.doi.org/10.1042/BSR20180289] [PMID: 29572387] ].
The more the noise or the poorer the image quality, the less precision the edge mapping process will have [12Chang HY, Jung CK, Woo JI, et al. Artificial Intelligence in Pathology. J Pathol Transl Med 2019; 53(1): 1-12. [http://dx.doi.org/10.4132/jptm.2018.12.16] [PMID: 30599506] ].
Further, the less accuracy on the border mapping process, the worse is the performance on the classification process [13Ware C. Perception for Design (Interactive Technologies) 2nd ed. 2004.].
In the literature, we did not find models that were used to recognize Candida spp. However, we found references bout the artificial neural networks to calculate antifungal activity against C. albicans. Furthermore, there are references to other cutaneous diseases [14Hogarty DT, Su JC, Phan K, et al. Artificial intelligence in dermatology-where we are and the way to the future: A review. Am J Clin Dermatol 2019. [http://dx.doi.org/10.1007/s40257-019-00462-6] [PMID: 31278649] -16Min S, Kong HJ, Yoon C, Kim HC, Suh DH. Development and evaluation of an automatic acne lesion detection program using digital image processing. Skin Res Technol 2013; 19(1): e423-32. [http://dx.doi.org/10.1111/j.1600-0846.2012.00660.x] [PMID: 22891680] ] such as melanoma [17Magalhaes C, Mendes J, Vardasca R. The role of AI classifiers in skin cancer images. Skin Res Technol 2019; 25(5): 750-7. [http://dx.doi.org/10.1111/srt.12713] [PMID: 31106913] ].
Hence, we established a goal. The objective was to reach a diagnostic accuracy of at least 80%. The model proposed by Manousaki et al. [18Manousaki AG, Manios AG, Tsompanaki EI, et al. A simple digital image processing system to aid in melanoma diagnosis in an everyday melanocytic skin lesion unit: A preliminary report. Int J Dermatol 2006; 45(4): 402-10. [http://dx.doi.org/10.1111/j.1365-4632.2006.02726.x] [PMID: 16650167] ] provided an accuracy of 89.4%. The model for pre-diagnostic digital imaging developed by Christensen et al. [19Christensen JH, Soerensen MB, Linghui Z, Chen S, Jensen MO. Pre-diagnostic digital imaging prediction model to discriminate between malignant melanoma and benign pigmented skin lesion. Skin Res Technol 2010; 16(1): 98-108. [http://dx.doi.org/10.1111/j.1600-0846.2009.00408.x] [PMID: 20384888] ] presented an accuracy of 77%. Our model is superior in accuracy to those mentioned above.
The final prediction model developed exhibits an accuracy of almost 90% and will thereby represent an improvement in diagnostic accuracy. Despite the different appearances and qualities of the images in the image database, we could develop a prediction model with accuracy better than that expected from an experienced dermatologist [20Ryan P, Luz S, Albert P, Vogel C, Normand C, Elwyn G. Using artificial intelligence to assess clinicians’ communication skills. BMJ 2019; 364: l161. [http://dx.doi.org/10.1136/bmj.l161] [PMID: 30659013] ].
The algorithms are thereby stable and not dependent on the quality of the image |under inspection. These new tools can also be utilized in telemedicine, where images can be uploaded to an automated web-based database and subsequently analyzed [21Kuziemsky C, Maeder AJ, John O, et al. Role of Artificial Intelligence within the Telehealth Domain. Yearb Med Inform 2019; 28(1): 35-40. [http://dx.doi.org/10.1055/s-0039-1677897] [PMID: 31022750] ].
We designed and programmed the “Candida identifier application” in Octave. It allowed the identification of the Candida species by locating certain geometric descriptors, such as the centroid and the surfaces of the circular objects that comprise the images. This system could differentiate the C. albicans from other varieties of Candida such as C. Kefyr, C. parapsilosis, and C. krusei, with accuracy. The analysis by the system of these descriptors allowed us to identify the Candida species in more than 90% of the cases. The sensitivity and specificity that were above 90% indicated a high diagnostic efficacy that was rather large to be the first version of a diagnostic program.
The application was highly precise in the identification based on the forms; however, it did not exhibit the same features in color-based identification. This element typically presents variations in the data, thus rendering identification difficult. We should acquire more copious databases to implement more precise identification methods, such as deep learning algorithms for images; however, the recognition of information in three different areas was satisfactory, as it allowed us to determine and classify injuries within the medical margins.
Finally, we emphasize that the objectives set at the beginning of the project were achieved. The results that we obtained from the “Candida identifier system” paved the way for handling images obtained from an optical microscope. This implies a wider range of materials can be used and specialized personnel economy (mainly in developing countries), and better service from the patient's perspective.
“Candydos” program. CRC designed and implemented the algorithm for “candydos” through Octave, as well as developed its program.
JLGC has contributed to create the algorithms of the program
RA has provided and selected Candida spp.
ETHICS APPROVAL AND CONSENT TO PARTICIPATE
This study was approved by the Ethics and Research Committee of Hospital General Manuel Gea Gonzalez, Estado
de México, Mexico. (Approval Number: OF /No 145001022 151/CCEEIS/016/2018).
HUMAN AND ANIMAL RIGHTS
No Animals were used in this research. All human research procedures followed were in accordance with the ethical standards of the committee responsible for human experimentation (institutional and national), and with the Helsinki Declaration of 1975, as revised in 2013.
CONSENT FOR PUBLICATION
Written informed consent was obtained from patients for the use of the fungal isolates and the publication of the cases details.
AVAILABILITY OF DATA AND MATERIAL
The data that support the findings of this study are available from the corresponding author [C.C] upon request.
The authors received no financial support for the research, authorship, and/or publication of this article.
CONFLICT OF INTEREST
The authors declare no conflict of interest, financial or otherwise.
Supplementary material is available on the publishers Website along with the published article.
Şular FL, Szekely E, Cristea VC, Dobreanu M. Invasive fungal infection in romania: Changing incidence and epidemiology during six years of surveillance in a tertiary hospital. Mycopathologia 2018; 183(6): 967-72. [http://dx.doi.org/10.1007/s11046-018-0293-2] [PMID: 30168077]
Aractingi S, Pellacani G. Computational neural network in melanocytic lesions diagnosis: Artificial intelligence to improve diagnosis in dermatology? Eur J Dermatol 2019; 29(S1): 4-7. [http://dx.doi.org/10.1684/ejd.2019.3538] [PMID: 31017580]
Wu M, Yan C, Liu H, Liu Q. Automatic classification of ovarian cancer types from cytological images using deep convolutional neural networks. Biosci Rep 2018; 38(3)BSR20180289 [http://dx.doi.org/10.1042/BSR20180289] [PMID: 29572387]