Blockchain-based Learning for Health Screening with Digital Anthropometry from Body Images

. Anthropoid images encode reliable biometric information in abundance. Recent research on image-based screening drives this effort to investigate the feasibility of interpreting the inherent nutritional state from multidimensional human body images. However, anthropometric databases are becoming increasingly essential and grow in parallel to achieve efficient system designs. This paper presents a novel imaging-based strategy in an augmented environment to quantify the human anthropometric features with blockchain-based learning to generate a diagnosis report. It includes evaluating the attributes such as height, weight, forearm, wrist, waist, Mid-Upper Arm Circumference (MUAC), Knee, Feet, head length from an image using the Augmented Reality and Blockchain-based Transfer Learning for diagnostic accuracy. We developed a novel skeleton known as FETTLE to determine the role of body measures for assessing nutritional conditions and body weight from human body images. It forms an instantly applicable technique aimed at evaluating children’s growth patterns all through their initial ages. FETTLE app can also be operated on bedridden people as a screening mechanism to spot their risk of pressure ulcers and undernutrition, followed by a more structured examination. Our approach is superior in accuracy measures with consortium blockchain-based learning context with privacy-preserved medical data sharing and high-end user experience and interaction. Our framework is proved to gain about 97.26% validation accuracy on anthropoid images.


Introduction
The accelerating influence of mobile gadgets and smartphones has significantly committed to the advancement of wellness screening applications. The present-day mobile and handheld devices carry high-end sensors furthermore computational strength for establishing an augmented mobile atmosphere. The usage of digital photogrammetry can be a viable alternative to the physical estimation of the human body in the medical field of research [1]. Few researchers [2] have addressed the question of information obtained by an image. A complete survey of imaging-based health screening tests is exhibited in the existing literature [3].
Usually, an image can provide implicit knowledge about the picture in terms of metric geometric relationships, as pictured in Fig.1. This technique can overcome constraints associated with the BMI calculation, precisely in the misclassification of obesity amidst the person with relatively high lean mass in bodybuilders or low lean mass in the elderly. Our approach to adopting digital images to assess human body form is predicated on evidence from earlier findings performed using either visual estimation or photographic assessment of body volume, from which body composition can be determined.

Fig. 1. Anthropometric classifications
We surveyed the FETTLE App's remarkable scope on accurately and scientifically measuring the human body composition by determining body size, structure, and balance. Furthermore, it assists psychologists to estimate the attributes like body size, height, arm length, etcetera in correlation with other human measurements, comprising sight (e.g., color, distance, and clarity), touch (e.g., sensitivity, weight, and pain), movement (e.g., rate and reaction time), memory, and mental fatigue [4]. There are requirements on a health-related application to serve as a personal physician to monitor individual health and development in the branch of Somatometry, Cosmetology, Cephalometry, Craniometry, and Osteometry by the identification of humans. Transforming the smartphone into a personal practitioner is what the FETTLE app is making all the way. It guides the users in knowing their health status and disease at risk by prompting their anthropometric values, thereby screening the overall wellness.
Several learning algorithms ease to infer medical information from health data. However, most machine learning algorithms are data-driven and require a training dataset from a vast number of body images and measures with heterogeneity. However, there are several real-world learning contexts where this hypothesis appears not to hold [5]. There are scenarios where the training dataset is cost-expensive or challenging to acquire. Hence, there is a necessity to design a high-performance learning model that can be trained with many easily gathered data from diverse areas. Thus, a methodology well-known as transfer learning has emerged. Its application-specific solution tends to be associated with the field of image processing and rendering. There are two stages in training a neural network of transfer learning techniques.
Initially, the neural network is supplied with a voluminous benchmark dataset with generally more labels to train in the pretraining stage. Subsequently, the trained network is extensively trained with target data with comparatively fewer labels in the later stage known as fine-tuning.
At the pretraining stage, the neural network is expected to be trained on the general features that can be applied to the target task to yield better performance than self-learning, where the network learns by itself. There is still considerable ambiguity concerning the impact of transfer learning on performance improvisation and the impact of the pretraining benchmark dataset, type, and size leaving open research questions for the medical image domain.
Blockchain is an emerging technology that finds its application in vast areas [6] because of its decentralized, peer-to-peer transaction and immutable nature. It is a growing data as a list of records, known as blocks, that are cryptographically linked to form chains [7]. The blocks are tied up with a cryptographic hash of the prior block, a timestamp, and transactions usually denoted as a Merkle tree. When machine learning and blockchain converge, the latter can benefit from the learning algorithm's ability to accelerate voluminous data analysis.
Leveraging the two together can feasibly emerge a novel paradigm.
Inline for patients suffering from mobility issues or when the patient is too sick to stand, unconscious, or bedridden, we are in a state to depend on their subjective approximations of body height and weight rather than acquiring an actual objective measure. This paper contests the claim that bedridden patients and born children need to have their body measurements calculated while they stay in bed. Several medical research cases proved that bed confinement or difficulties in sustaining an erect and upright position, inappropriate material, lack of knowledge, and time are well-known limitations that affect the systematic body measurements in hospitalized patients.
The remainder of this paper is structured as follows: section 2 elaborates on the research works related to our problem statement, followed by section 3 that provides a system overview. Highlevel Architecture is detailed in section 4. The fifth chapter is concerned with the methodology used for this study. Observed effects are presented in section 6, whereas Accuracy and performance analysis in sections 7 & 8. Finally, the conclusion gives a summary and critique of the findings paving the way to the identification of areas for further research.

Related Work
The human attributes quantification procedure from an image in a two-dimensional illustration is described in previous researches [8]. The human body fat calculation method is beneficial in generating the dataset for weight prediction [9]. It projects a computational core to process both single two-dimensional images or a pair of 2D models to examine body weight and BMI. This A real-time approach to incremental scene perception with mobile platforms reveals how AR is rendered in the camera scene [15]. ARKit is capable of detecting a maximum of 100 images simultaneously to automatically estimate the physical size of the object in the captured image.
It is capable of generating three-dimensional mesh data from the screen geometry [16]. The method of deducing standing human bodies from single images using the calibration models is presented in the previous works [17]. The purpose of the researcher's inquiry in [18]  and estimates the segmentation results [19]. Another approach on age synthesis and estimation via facial images describes human face recognition concerning age and other factors [20]. The literature [21] tends to use a methodology to refer to 3D facial landmark localization with regression from extracted features. It details the landmark localization of the human body shape triggering it with an asymmetric design, thereby applying shape regression. The other strategy uses kinects in scanning 3D full human body images to perform the full-body scan and extract the features as required [22]. Typical study on privacy-preserving cloth trial with Mobile Augmented Reality aids in bringing out the cloth try-on mode of picture rendering to size attribute [23] [25]. The method for extracting facial characteristics by 3D facial landmark localization [24] by asymmetric patterns [27] and shape regression [26] from deficient local features has been proposed [28]. Experts have always seen 3D imaging as a rich source of age estimation and synthesis [30] [33]. Motivated by the research works [31][32], blockchain can share data and learning models in a deep learning framework to enhance transfer learning in a decentralized and distributed environment. With the help of learning strategies to govern blockchain, there is also a chance to enhance security significantly.
Further, as learning algorithms love to work with many data, it creates an opportunity to build better models by taking the benefits of blockchains' decentralized nature that encourages data exchanges. Health apps [29] [34] typically aid two types of functions. The primary task is to collect or store health-related records, which some applications enable the user to share with a health care provider. Another task is access to health information, such as nutrition data on balanced foods, healthy diets, and workout routines. However, exclusive early efforts required vital labor to process the images and figure out the body mass. Using automated digital image analysis, the computerized visualization method might overcome some of the problems and the visual estimate of body volume and consistency. Therefore, this research aims to promote an easily accessible, compact, agile, and comparatively economical but efficient computerized image interpretation with the blockchain-oriented learning process for large-scale estimation of nutritional deficiency in its more initial stage.  permits an innovative prospect of incentivizing smart contracts. Table 2 lists the notations and symbols used throughout the manuscript.

FETTLE: System Overview
In general, ARKit is the framework that renders scene-capturing features for image processing and transcription. FETTLE is an ARKit application that enables the back and front camera of the device to image the detected special points by locating any number of 3D objects in space.
It performs vector operations to calculate the distance and direction of the camera from a point in imaging space to assess the volume of the object despite the movement of the camera holder.
This data is later used to generate the control points. Further, the application quantifies the distance between the special points and holds back the statistical information about the special points in a detected plane to a file to generate a 2D or 3D content. A significant feature of this app is model creation that facilitates the construction of a visual object for the effectual constitution of 3D data. The generation of a 3D model from 2D or 3D health images integrates in real-time to turn it to be more interactive and effective application.
The human body images are captured from a mobile camera or smartphone in an augmented context. It needs to be pre-processed by cropping the detected shape from the original image to fix the control points. The coordinates of control points are used for the extraction and computation of anthropometric measures. We employ the Convolutional Neural Networks validation data are procured from digital camera viewpoints, conventional machine learning procedures can predict results better. However, when the training data is of digital camera views, and the validation data is from anthropometric characteristics, then the prediction effects are apparent to deteriorate due to the data variance in the domain. An alternative means to inspect the data domains in a transfer learning context is that the training data and the validation data are diverse sub-domains connected by a common high-level area. Hence, we need transfer learning, as there is a confined quantity of target training data. The detailed system overview is described in the subsequent subsections, which covers the entire concept of delivering a valuable app for estimating an individual's health screening. The system insights are diagrammatically presented in Fig. 2.

Adulthood Space
The Adulthood Sector of the app furnishes the digital anthropometry of an adult to screen their health risk and analysis, if any. A diagnostic health screening from a human body image with virtual measures is one of the novelties of our approach.
The stature Discoverer section initiates the camera to capture the photo of a person subjected to the AR Scene to determine the height. It further instructs the users on their position to capture a quality image worth measuring an accurate height by standing barefooted on a hard, flat surface with back alongside a vertical plane and feet be set apart just less than shoulder-width.
By setting up the control points on the image, as shown in Fig.3, the app shares the information left to the user to manipulate how to place the control points for a height discovery.
where, ω = BMR value, δ = weight, λ = height and γ = age The athlete Space module is useful for an individual in terms of the athlete world. The fitness report in playing is the motto of this module. FETTLE helps athletes to monitor their Fitness status by the anthropometric values obtained from their image.

Baby Care Space
The Baby Care Space is developed to determine the baby's anthropometry from an image. This section acts as a Baby Nutritionist in assessing the Baby growth and deficiencies. Results from this section are helpful to identify whether the baby is developing typically or not.

Immobilized Patients Anthropometry
This module is specially developed for bedridden patients to determine their anthropometry.
Bedridden patients wing In this section, hospitalized patients were subjected to anthropometry, which is easier to monitor regularly. Measurement of height within the critical care unit is essential for approximating ideal body weight. FETTLE provides a solution in estimating the height and weight of bedridden patients. Yet another possibility is to identify the spread, and severity to diagnose the stages of pressure ulcers (bedsores) from digital images.
Height determination In this ideality, our novel approach is to obtain the height of the bedridden patients from an image using the below formula where λ is the height of the patient, β is the knee height of the patient, and γ is the age of the patient.

Health Trends
This section exclusively assists the app users to keep hold of a health record to track their health status based on the overall anthropometric results acquired with the current health trends. A comprehensive report of an individual user is provided with visual trends. We have employed an exclusive blockchain network that aids in learning by extending the medical data/dataset privately shared.

High-Level Implementation Architecture
A high-level system architecture diagram is detailed in the below Fig.5 User log in to the app; new user registration is also available; thereby, the user gains access to the FETTLE App.
Once login is the successful user enters a health screening section. The user interfaces consent in drawing out the data from the virtual environment in proportion to the availability and accessibility of the objects. The augmented world has options for arriving at the report as well     The incentivization uses ethers as cryptocurrency by leveraging the underline Ethereum base, used in data sharing from one user to another to respond to service or resource utilization. If a user has a large dataset and can be given medical diagnostics, the user can automatically rend that dataset for monetary benefits by sharing it using a private channel. However, the pseudonymity is preserved as it is private data, and the scheme is elaborated in Algorithm 2.
The provider can ask for a certain amount of ethers for data exchange to use the dataset by some other users. The overall sequence of actions that can be performed with the system is

Mathematical Model
We consider a source domain space Ɗ and target domain space Ɗ represented by twin tuple with instance space (Ƶ) and hypothesis space (ѡ) enclosed. The objective function can be defined as the marginal probability between the label space (Ƚ) and the hypothesis space and is denoted as ɳ.
The source and target constraints differ in three feasible ways in our case. 3. There is adrift in the conditional probability distribution of source and target domains are distinct.
The transfer learning is a mapping function that randomly maps training data z ∈ Ƶ to a hypothesis ∈ ѡ defined by a conditional distribution, P (Ƚ | ѡ ) with the information gained from the source domain space in consort with the blockchain collected peer medical data (Ɓ ) implied by Info-Gain (Ɗ + Ɓ ). Hence our testing implies that this approach can also detect other undernutrition baby diseases. Table 2 depicts the performance of the sharing process supplied by the blockchain module to the framework.

Smart contract creation 18045
Smart contract update 2039

Raft block time 50
Transaction processing time 2028 Fig. 9. The output of Turi Create Visualization

Performance Analysis
Our model improves its accuracy by employing the following constraints.
 Applying Transfer Learning  Procuring additional data for training  Using a more complex, deeper (more layers) CNN It is apparent in the chart below (Fig 10) that the FETTLE app, when made to learn by self with data and deep learning architecture, achieved a relatively lower accuracy when the transfer learning module is done to equipped so that the model learned from the experience of others with higher accuracy.  Table 3, and Table 4 shows the R-squared value of the models.      Image Augmentation and Hyper-Parameter Tuning methods are employed to yield better performance of our model.
 Image pre-processing technique like rescaling is followed to ensure that the training images are appropriate for accuracy.
 Usage of Turi Create produces a reasonable model size, optimized, and specialized for iOS Mobile environment. Table 5 lists the factors by which we determine the overall performance of our FETTLE App. Table 5. Performance analysis of FETTLE.

Calculations
The test measured the time to calculate the number n to ten thousand decimal places, for which our app is returning a better result.

Network Support
This test measures the reachability of the network in both the online and offline modes of the app. FETTLE yield a good result even in this experiment.

Image Training Speed
Training speed is an important test to be carried out since many of our modules are handled with images. Since we have used ResNet CNN as an image classifier, the overall app size is pretty good compared to other classifiers.
Computational power FETTLE App yields a good result in computational power since it uses most of the iOS Native plugins and Frameworks. The use of ARKIT and COREML is one such example to obtain better performance.
Abbreviations and Acronyms: FETTLE App define abbreviations and acronyms the first time they are used in the text, even after they have been defined in the abstract.

Concluding Remarks
This work examines the relationship between human body composition and visual body appearance to measure health from multidimensional body pictures. We presented a deep convolution neural network as a transfer learning methodology for identifying human anthropometry scientifically. We proposed a module that provides a decentralized blockchainoriented learning strategy. The primary motivation behind FETTLE App was the need to design an efficient algorithm for determining anthropometry to assess health screening with the aid of mobile cameras. A digital health guide strives for each of us to identify the health risks proactively before seeking a medical practitioner. Our approach is guaranteed to render a health estimate from body images visually based on all investigational outcomes. As the future scope of the work, we are trying to share the training model among the peers and to lend a dataset on a timely basis. Forensic anthropometry can be developed for detecting criminals by comparing the fed criminal photographs into FETTLE.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have influenced the work reported in this paper.