A fuzzy rule based machine intelligence model for cherry red spot disease detection of human eyes in IoMT

Internet of medical things (IoMT) plays an important role nowadays to support healthcare system. The hospital equipment’s called as medical things are now connected to the cloud for getting many useful services. The data generated from the equipments are sent to the cloud for getting the desired service. In current scenario, most hospitals collect many images using equipments, but these equipments have less computational capability to process the huge generated data. In this work, one such equipment is considered which can take the human eye images and send the images to the cloud for detection of cherry red spot (CRS). CRS disease in eyes is considered as one of the very dangerous disease. The early diagnosis of CRS disease needs to be focused in order to avoid any adverse effect on human body. In this paper, a machine intelligence based model is proposed to detect the CRS disease areas in the human eyes by analyzing several CRS disease images using IoMT. The proposed approach is mainly focused on fuzzy rule-based mechanism to carry out the identification of such affected area in the eyes in cloud layer. From the results, it is observed that the CRS disease areas in the eyes are detected well with better detection accuracy and lower detection error than k-means algorithm. This approach will help the doctors to track the exact position of the affected areas in the eye for its diagnosis. The simulation is performed using socket programming written in Python 3 where a cloud server and a client device are created and images are sent from the client device to the server, and afterwards the detection of CRS is performed at the server using MATLAB R2015b. The proposed method is able to provide better performance in terms of detection accuracy, detection error and processing time as 94.67%, 5.33% and 1.1481% units respectively on an average case scenario.


Introduction
Nowadays, healthcare technology is a major area of research where human life is considered as top priority [1,2]. The IoMT is such a healthcare technology that comprises of medical things or in other terms medical equipments/devices which are connected to each other to the cloud with high bandwidth to provide services to the users [3][4][5]. It is playing vital role in healthcare industry where hospitals, medical sensors, medical equipment, other devices are connected to each other and to the cloud. The IoMT has many applications such as monitoring health of a person in periodic manner, medical data storage, faster data streaming, alarming systems, ambulance service, advancing healthcare technology, faster computation service using AI, telemedicine, medical robotics etc. [6][7][8]. There are many hospitals which have huge number of medical sensors and equipment's, so if we consider a city then the data generated from a hospital by using the medical devices for its large number of patients is huge [9]. If the generated data is stored in a central database server then it can be used for telemedicine and advanced healthcare applications. Therefore, IoMT is a booming area of research to deal with all type of healthcare problems and facilities.
In this work we consider a use-case where hospital uses its eye image capturing device that detects the CRS disease at the cloud using IoMT system. CRS disease is considered as a lipid storage disorder as well as fatal genetic disorder in the macula of the eye which causes central retinal artery Kalyan Kumar Jena, Sourav Kumar Bhoi, Debasis Mohapatra, Chittaranjan Mallick, Kshira Sagar Sahoo and Anand Nayyar contributed equally to this work.
Extended author information available on the last page of the article occlusion and progressive damage of the central nervous system [10][11][12]. It arises due to the blockage of the blood flow through the central retinal artery. This disease can also affect the mental as well as physical abilities of human beings. It mainly occurs due to the absence of an essential enzyme Hexosaminidase-A and more quantities of gangliosides accumulated in the nerve cells of brain. So, it is much essential to track the exact shape, size and location of this disease to carry out the diagnosis in a better way. It is very much essential to focus on the early diagnosis of this disease in order to avoid any adverse effects on the human body.
Machine Intelligence (MI) [13] plays an important role in the health care system in the current scenario. Fuzzy rule based mechanism [14] can also be considered as an important prospective of MI for the analysis of health care data. Fuzzy rule based mechanism can be used to predict the output by analyzing several input features. So, both MI and Fuzzy rule based approach can be used as a solution for the analysis of several patient data in the health care system which can accelerate the process of diagnosis [13][14][15].
The accurate detection of CRS infected area is a challenging task in the current scenario which needs high end computational resources at hospitals for accurate detection. MI as well as Fuzzy rule-based mechanism can be considered as the solution for such detection which can be provided on demand using cloud service. So, in this paper, we have focused on MI based on Fuzzy rule based model for the detection of CRS disease from the analysis of several CRS disease images. This work is mainly focused on the recognition of CRS disease affected regions of human eyes. The proposed work is able to track the CRS disease affected areas with better detection and lower detection error. This work can help the doctors in detecting the CRS infected area by analysing several CRS disease images in the real world scenario. The main contributions of this work are stated as follows.
• An MI based model is proposed for the detection of CRS disease region in the human eyes from the analysis of several CRS disease images at the cloud using IoMT platform. • The proposed approach is mainly focused on fuzzy rule based mechanism for the identification of such affected region in the human eyes. • The proposed method is able to track the CRS disease areas in the eyes with better detection accuracy and lower detection error than k-means algorithm. • The simulation of this work is carried out using Socket programming and MATLAB R2015b and evaluated using the performance parameters such as detection accuracy and detection error.
• The proposed work is able to provide better performance in terms of detection accuracy, detection error and processing time as 94.67%, 5.33% and 1.1481% units respectively on an average case scenario.
The rest of the paper is organized as follows. Section 2 describes the related work, Sect. 3 discusses the proposed model, and Sect. 4 describes the analysis of the proposed model. The results and discussion have made in Sect. 5 and finally Sect. 6 concludes this work.

Related work
Different works have been carried out by different researchers related to CRS disease. Some of the works are described as follows. Varela et al. [1] focused on the macular CRS multimodal ophthalmic imaging by using a sialidosis type I cohort and a quantitative mechanism. Alzebaidi et al. [16] emphasized on progressive familial intrahepatic cholestasis type I which is associated with CRS. Riboldi et al. [2] discussed on the identification of CRS myoclonus in the eye. In [17], discussed on the detail investigation on type I sialidosis, CRS and its processing mechanism. Fukuyo et al. [3] emphasized on the analysis of galactosialidosis Type IIb with bilateral macular CRS but mild dysfunction. In another work, Soleimani et al. [4] emphasized on the analysis of CRS myoclonus syndrome. Researchers in [18], provides the detail investigation report of CRS in a patient with Tay-Sachs disease. In [19], authors emphasized on the analysis of CRS and ocular manifestations of sphingolipid-mediated neurodegenerative and inflammatory disorders. Nakaya-Onishi et al. [10] have implemented the time course variations of the CRS in a patient with Tay-Sachs disease. Heroman et al. analysed CRS in sialidosis specifically in type I mucolipidosis [11].
In [12], authors provide a detail investigation report of sialidosis by focusing on fundus autofluorescence and optical coherence tomography of a macular CRS. Another investigation report designed by [20], which focuses on macular CRS which can help diagnose the exceptional storage disorder in a child with frequent respiratory region infections. In [21], Federici et al. analysed bilateral orbital infarction, ophthalmic artery occlusion and CRS. Recognition of two novel mutations in neuraminidase-1 (NEU1) gene by focusing on multigene panel next generation sequencing in a patient with CRS emphasized in [22]. From the above analysis, it is observed that different works have been carried out by different researchers and scientists related to CRS. However, accurate identification and classification of infected region as well as adoption of processing mechanism is a challenging issue in the current scenario. The summary of different methods is mentioned in Table 1.

Proposed model
The proposed model is described in Fig. 1. This model describes the entire process for the analysis of several CRS disease images to track the infected area in a better way. The proposed framework mainly consists of two layers such as device layer and cloud layer. In the device layer, there are so many hospitals which have eye image capturing equipments/devices. Initially, all the CRS disease images taken by the equipment are sent to the local server. The local server is maintained by the hospital to send and receive the CRS disease images and its relevant information. The local server is not so much efficient to process all the CRS disease images due to lack of computation facilities as well as lack of sufficient storage space. So, from the local server, the CRS disease images will be sent to the cloud server through the gateway to carry out the analysis process. The proposed method which is focused on Fuzzy Detail investigation report of CRS in a patient with Tay-Sachs disease The accumulation of lipid in retinal ganglion cells that leads to a chalk-white appearance of the fundus called as CRS is the hallmark of Tay-Sachs disease Chen et al. [19] Analysis of CRS and ocular manifestations of neurodegenerative and inflammatory disorders Ocular inflammatory and autoimmune diseases involve activation and migration of endothelial cells, neovascularization, and egress and infiltration of immune cells to the uvea, anterior chamber, and cornea Nakaya-Onishi et al. [10] Time course variations of the CRS in a patient with Tay-Sachs disease CRS developed and then diminished in both eyes during 5 years and 8 months period Heroman et al. [11] Analysis of CRS in sialidosis specifically in type I mucolipidosis Ophthalmoscopic variations observed in the patient due to an accumulation of sialyloligosaccharides in the inner layers of the retina Zou et al. [12] Sialidosis with fundus autofluorescence and optical coherence tomography of a macular CRS Ophthalmic images using fundus autofluorescence and optical coherence tomographyare are considered as useful methods to detect neurological metabolic disorders dealing with a CRS in the macula Padhi et al. [20] Macular CRS in diagnose the exceptional storage disorder in a child with frequent respiratory region infections Fundus examination showed bilateral CRS at the macula Diffusely distributed hyper-pigmented patches on the back and extensor aspect of the extremities Federici et al. [21] Analysis of bilateral orbital infarction, ophthalmic artery occlusion and CRS Ophthalmic artery occlusion produces profound vision loss due to the simultaneous ischemia of the choroid and retina. A CRS can be an acute initial sign but is more suggestive of central retinal artery occlusion Mutze et at. [22] Recognition of mutations in NEU1 gene by focusing on multigene panel next generation sequencing in a patient with CRS rule-based MI approach will be implemented in the cloud layer where the server processes several CRS diseases images to detect the disease. The proposed approach will be applied on the CRS disease images to track the infected region with better detection accuracy and lower error rate. After obtaining the infected region from the analysis of several CRS disease images, it will sent for various applications or further processing to accelerate the diagnosis process in a better way. The proposed approach can be considered as a solution for the identification of CRS disease region from the analysis of several CRS disease images to accelerate the diagnosis process. In this work, we have mainly focused on the application of Fuzzy rule-based MI approach for the identification of infected region by analysing several CRS diseases images. The fuzzy rule-based approach is based on the computation of degrees of truth. It works on the levels of possibilities of input to explore the definite output. It can able to solve a problem after consideration of all available data. Afterward, it can able to take the best possible decision for the given input.
The proposed methodology is mentioned in Fig. 2. In this work, the triangular membership function is applied on the CRS infected image to compute the membership value (mv) value for red, green, and blue planes. Then the Fuzzy rule-based approach is applied to find the CRS infected region. The proposed method is described with the help of five algorithms such as Algorithms 1 to 5. The proposed method is compared with the k-means method in terms of detection accuracy and detection error. The k-means algorithm is a clustering algorithm that is considered as an unsupervised machine learning algorithm. The k-means algorithm is used to group similar items in the form of clusters. It attempts to partition the dataset into k pre-defined distinct non-overlapping subgroups or clusters where each data point belongs to only one group. It can segment the interest area from the background of the image. This algorithm is performing the grouping mechanism in data in such a way that the items in the same cluster are more similar to each other than those from different clusters. The k-means algorithm is mentioned in Algorithm 6.
In this work, we have used triangular membership function for the detection of infected region in the human eyes and it is described in Algorithm 1. This function is used to generate the mv by considering the parameters i, j, k with the condition i\j\k. This function is computed by focusing on the value (v). The mv value is considered as 0 when v value is less or equal to i or the value of k is less than or equal to v. The mv value is considered as ðv À iÞ=ðj À iÞ when the v value lies in between i and j. The mv value is considered as ðk À vÞ=ðk À jÞ when the v value lies in between j and k. The discussed principles are used in our work for the processing of CRS disease images.  The Algorithm 1 is used for the computation of mv for Red (Re), Green (Gn) and Blue (Bl) plane. The mv computation mechanism for Re, Gn plane is mentioned in Algorithms 2 and 3 respectively. It has been classified the mv values into four regions such as mv low (mvL), mv medium (mvM), mv high (mvH) and mv very high (mvVh) by focusing of uniform distribution mechanism. From the Algorithm 2, the mvL will be considered as Re/31, if the Re value lies between 0 to 31. The mvL has considered as ð63 À ReÞ=31, if the Re value lies between 32 to 63. The mvM has taken as ðRe À 64Þ=31, if the Re value lies between 64 to 95. In the similar way, the mvVh will be treated as ð255 À ReÞ=31 , if the Re value lies between 224 and 255.
The Algorithm 3 is similar to Algorithm 2. The mvL will be considered as Gn/31, if the Gn value lies between 0 to 31. The mvL will be considered as ð63 À GnÞ=31, if the Gn value lies between 32 to 63. Similar to this, the mvVh will be considered as ð255 À GnÞ=31, if the Gn value lies between 224 to 255. In the same way, we can design the computation module for blue plane.
In this work, it has been classified the entire image region into two parts such as high active (HA) region and low active (LA) region. The infected region is represented using HA region in our work. Here, we have represented the HA region using red color (RC) and LA region using black color (BC). Algorithm 4 describes the identification of region status (RS) as HA or LA and color status (CS) as RC or BC by considering the several color combination (CC) of Re, Gn and Bl color. For instance, the RS and CS are considered as LA and BC respectively if we consider the CC as Re ¼¼ Vh&&Gn ¼¼ M&&Bl ¼¼ L or Re ¼¼ Vh&&Gn ¼¼ H&&Bl ¼¼ H and so on. We have taken the mentioned CC by focusing on a standard RGB color model (CM) [14].
Algorithm 5 describes the entire process of infected region identification by analysing several CRS disease images. After taking the CRS disease image, the mv values for Re, Gn and Bl plane are computed by considering the processing mechanism as mentioned in Algorithms 1 to 3. The RS and CS are decided by considering the processing mechanism as mentioned in Algorithm 4. All algorithms are implemented using MATLAB R2015b [23].

Analysis of proposed model
In this section, we described about the computational complexity of the proposed model. Here, computational complexity mainly describes about the time and memory complexity at the cloud after the CRS images of a patient are sent through the IoMT devices to the cloud device. Let, an image is considered as r number of rows and c number of columns, then by using proposed fuzzy rule based approach at the cloud, the membership of the pixels ðr i ; c j Þ of an image takes a total time of Oðr Â cÞ. Similarly, the memory usage for finding the membership of the pixels for the image takes total space complexity of Oðr Â cÞ.
As per system requirement, for diagnosis of CRS disease of a patient, let n number of same dimensional images are required for detecting the disease. The majority of the results are considered to detect the CRS disease more accurately. For processing n number of images of same dimension at the cloud, a computational complexity of Oðn Â r Â cÞ is needed. For detecting the actual CRS affected pixels, majority of the pixel locations are considered from n number of images to validate the areas. The majority is decided based on a threshold value t which is decided as per the system requirement. However, in the simulation section we have only considered a single image to be sent to the cloud for detection of the CRS disease using the fuzzy rule based approach. Here, in the simulation section many different CRS disease images with different dimensions are taken to validate the work by showing the accuracy and error.
As per the proposed model, we have some assumptions as follows. The images are not processed at the local IoMT device as it lacks MI model and that service is provided by a cloud service provider to detect the CRS disease of a patient more accurately. The service provider here uses proposed fuzzy rule based approach to collect the n same dimension images of a patient and validate the actual affected area. Here, we assume the hospital IoMT devices are connected to the cloud with high bandwidth and sending the required number of CRS images of a patient to the cloud is not an issue. Therefore, the time complexity of sending the images to the cloud and receiving the CRS detected image to the hospital is out of scope of this work.
Here, we have only presented a framework and shown the MI part at the cloud level.

Result and discussions
In this work, we have considered several CRS disease images with different size from the source [23][24][25][26][27][28][29][30][31][32]. These images are mentioned in Figs. 3, 4 Table 2. The simulation of this work is carried out using Python Socket programming in Anaconda Jupyter platform and MATLAB R2015b [23]. For which, using socket programming a cloud server was created with ip: 127.0.0.1 and port: 8080 and one client is created in the same machine and from the client (local server machine) the CRS images taken above are sent to the cloud server to show the IoMT platform. Afterwards, the images received at the server are processed using MATLAB tool using the proposed methodology to detect the CRS disease areas. The proposed method is compared with k-means method by using the performance parameters such as detection accuracy (in %) and detection error (in %). These performance parameters are described as follows.
Detection accuracy It is referred to as how accurately the CRS infected region is detected. In this work, it is expressed in percentage.
Detection error It is referred to as the percentage of error arise during the detection of CRS infected region.
In this work, the computation time (processing time) for the processing of CRS images by applying the proposed method and the k-means method are described in Fig. 12. Figure 10 describes about the detection accuracy after processing the CRS disease images. Figure 11 illustrates the detection error (in %) after processing the CRS disease images. The detection accuracy, detection error and processing time computations are carried out using MATLAB R2015b.
The proposed method is able to detect the infected region with better detection accuracy with considerably lower detection error rate. The detection accuracy (in %) for the CRS disease images 1-7 [mentioned in Figs. 3 Fig. 10. Similarly, the detection error (in %) for the same set of CRS images using the proposed method are computed as 3.27, 2.20, 2.85, 2.62, 1.41, 1.46 and 1.99 respectively. The proposed method used for detection of the CSR infected region shows less detection error as compared to k-means method as shown in Fig. 11. The detection accuracy as well as the detection error computation varies from image to image. In this work, the detection accuracy for the CRS image mentioned in Fig. 3(a) is computed as 96.73% which is lower as compared to the results of other CRS images. The detection accuracy for the CRS image mentioned in Fig. 7(a) is computed as 98.59% which is higher as compared to the results of other CRS images. This scenario is  Fig. 10. The scenario of detection error is visualized in Fig. 11. From the above analysis, it observed that, the proposed approach is able to provide the detection accuracy which is always above 96% and the detection error rate is very low (maximum error is less than 3.5%) in this scenario. The proposed approach is able to provide maximum detection accuracy of 98.59% for the CRS disease image 5 mentioned in Fig. 7(a). The proposed approach is able to provide the detection accuracy which is above 95% after processing of all images in this scenario. From the results, it is observed that in some cases, the processing time for each CRS image does not depend on the size in this scenario. The processing time for the CRS image mentioned in Fig. 6(a) (size: 625 9 512) is computed as 0.3744 unit, however the processing time for the CRS image mentioned in Fig. 7(a) (size: 326 9 278) is computed as 0.8112 unit which is higher as compared to the processing time of CRS image mentioned in Fig. 6(a). This scenario is illustrated in Fig. 12. From the experiments, it is observed that the proposed approach is able to recognize the infected region from the analysis of CRS disease images in a better way as compared to k-means method in terms of detection accuracy and detection error. The proposed method is able to process each CRS disease image in a faster way. The processing time (in unit) for the CRS disease images using the proposed method are  Table 2 represent the average detection accuracy, detection error and processing time. From Table 2, it is observed that the proposed method provides the higher detection accuracy (in %, detection error (in %) and processing time (in units). This method is able to provide 3.079% better performance in terms of detection accuracy and 0.0035 units better performance in terms of processing time as compared to k-means method on an average case scenario.

Conclusion
This paper proposed an MI-based model to detect the CSR disease affected areas in the human eyes by analyzing several CRS diseases images in the cloud using the IoMT platform. In this work, a fuzzy rule-based approach is proposed to carry out the identification of CRS disease areas in the human eyes. From the results, it is concluded that the proposed method can identify the CSR disease areas in the eyes in terms of better detection accuracy and lower detection error from the analysis of several CRS disease images of different sizes. With this approach, the detection error rate is comparatively low as compared to k-means method. This approach will help the doctors to recognize the exact region of the CRS disease in the eye which will lead to the diagnosis of this disease in a better way. The proposed work is evaluated using Socket programming and MATLAB R2015b and found to be better in terms of better detection accuracy and lower detection error than k-means. However, it can provide better detection accuracy, detection error, and processing time as 94.67%, 5.33%, and 1.1481 units respectively on an average as compared to the k-means method. This work can be extended to implement the proposed methodology for the detection of red leaf spot disease from the analysis of several red leaf spot disease images. Further, it can also be focused on the development of enhanced methods to gain improved detecting accuracy with lower detection error and lower processing time.