Smart Diagnostic System for Blood Vessel Blockage Using Non-Invasive Methods (Without Angiography)

Smart Diagnostic System for Blood Vessel Blockage Using Non-Invasive Methods (Without Angiography)

Nikita Gosavi

Department of Computer Engineering JSPMs JSCOE, Pune

Samarth Ghorpade

Department of Computer Engineering JSPMs JSCOE, Pune

Varun Chaudhari

Department of Computer Engineering JSPMs JSCOE, Pune

Nayan Pawar

Department of Computer Engineering JSPMs JSCOE, Pune

Vaibhav Raul

Department of Computer Engineering JSPMs JSCOE, Pune

Abstract – Cardiovascular disease is still the cause of death globally. Sudden heart problems happen because of blocked blood vessels usually due to atherosclerosis. The best tests we have now like angiography are invasive use a lot of radiation are costly and not suitable for regular check-ups, especially in rural areas where medical resources are limited.

This study introduces a non-invasive diagnostic system that uses AI. It combines Photoplethysmography (PPG) infrared imaging and Doppler ultrasound with a deep learning model hosted in the cloud. The system collects data from sensors in real-time cleans it up to remove errors and sends it securely to the cloud for analysis to detect blockages.

The system we propose is designed to be affordable, able to handle a lot of users and easy to use in healthcare settings. By using -invasive technology, AI and cloud computing it aims to help with early heart screenings make diagnoses more accessible and reduce the need, for invasive tests.

Index TermsNon-Invasive Diagnostics, Blood Vessel Block- age, Cardiovascular Disease, Photoplethysmography, Doppler Ultrasound, Thermal Imaging, ensemble Deep Learning, AI in Healthcare.

1. Introduction

   Cardiovascular disease is one of the reasons people die all around the world. When we have atherosclerosis it slowly makes our arteries smaller. Reduces blood ow to important parts of our body like the heart and brain. If we can nd out about blocked arteries early we can do something about it. Lower the risk of death.

   Now the best way to nd out if we have a blockage is with a coronary angiography. But this method is not very nice because it is invasive it costs a lot of money. It is not good for people who live in rural areas or places with limited resources. We have made some progress with ways to sense what is going on in our body store information on the internet and use computers to learn from data. These new developments give us a chance to make a system that can diagnose problems without hurting us. This paper is about a system that uses

   devices like PPG, thermal imaging and Doppler ultrasound along, with a type of computer program called ensemble to detect blockages in our arteries. The system is called a Smart Diagnostic System. It uses deep learning to gure out if we have any blockages. Cardiovascular disease and blockages are the focus of this Smart Diagnostic System.

   latex

2. LITERATURE REVIEW

   1. Physiological Basis of Vascular Blockage

      Blood vessel blockage primarily occurs due to atheroscle- rosis, a condition in which cholesterol, calcium, and fatty substances accumulate along arterial walls. These deposits gradually narrow blood vessels and restrict blood ow to vital organs such as the heart and brain. Reduced blood circulation can result in ischemia, myocardial infarction, stroke, and other severe cardiovascular complications. According to the World Health Organization (WHO), cardiovascular diseases are re- sponsible for nearly 17.9 million deaths annually worldwide [1].

      Hemodynamic changes caused by arterial stenosis affect blood velocity, arterial elasticity, pulse wave morphology, and peripheral perfusion. These physiological changes can be observed using non-invasive sensing modalities such as Pho- toplethysmography (PPG), Doppler ultrasound, and infrared thermal imaging. The combination of these sensing techniques enables early detection of abnormal vascular conditions before complete blockage occurs.

   2. Photoplethysmography (PPG)

      Photoplethysmography is a non-invasive optical sensing technique used to measure changes in blood volume during the cardiac cycle. A typical PPG sensor consists of a light source and a photodetector that capture variations in reected or transmitted light through biological tissues. These variations

      produce pulse waveforms that contain valuable cardiovascular information [2].

      Researchers have shown that features extracted from PPG signals, including pulse transit time, augmentation index, sys- tolic peak amplitude, and pulse width, are closely associated with arterial stiffness and vascular abnormalities. Owing to their low cost and continuous monitoring capability, PPG sen- sors are widely employed in wearable devices and healthcare applications [3].

      Several studies have demonstrated that machine learning and deep learning algorithms can improve cardiovascular disease prediction using PPG signals. Signal preprocessing techniques such as ltering, normalization, and motion artifact removal signicantly enhance diagnostic performance.

   3. Infrared Thermal Imaging

      Infrared thermal imaging is a non-contact diagnostic method that measures temperature distribution patterns on the human body. Since blood circulation inuences skin temperature, regions affected by vascular blockage often exhibit abnormal thermal patterns compared to healthy tissues [4].

      Thermal imaging has found applications in detecting di- abetic foot ulcers, breast cancer, inammation, and periph- eral vascular disorders. Recent advances in thermal camera technology have enabled the development of portable and affordable thermal imaging systems suitable for healthcare applications.

      Articial intelligence techniques integrated with thermal imaging improve diagnostic accuracy by automatically extract- ing discriminative temperature features. Convolutional Neural Networks (CNNs) have demonstrated promising results in medical image classication tasks and thermal pattern analysis.

   4. Doppler Ultrasound

      Doppler ultrasound is a widely used non-invasive technique for measuring blood ow velocity using the Doppler effect. Frequency shifts generated by moving blood cells are analyzed to evaluate vascular conditions and identify arterial narrowing [5].

      Clinical parameters such as peak systolic velocity, end- diastolic velocity, pulsatility index, and resistive index are commonly used to assess vascular abnormalities. Doppler ultrasound has been extensively applied in the diagnosis of carotid artery disease and peripheral arterial disorders.

      Recent technological developments have enabled compact Doppler sensors to be integrated with embedded systems and cloud platforms. Automated signal analysis using articial intelligence reduces dependence on expert interpretation and improves diagnostic efciency.

   5. Deep Learning for Cardiovascular Diagnostics

      Deep learning has transformed healthcare diagnostics by en- abling automatic feature extraction from physiological signals and medical images. Convolutional Neural Networks (CNNs) have shown remarkable performance in image classication and pattern recognition due to their hierarchical feature learn- ing capabilities [6].

      Among different CNN architectures, MLP And CNN is widely adopted in medical image analysis because of its simple architecture, transfer learning capability, and robust feature extraction performance. Researchers have successfully employd deep learning techniques in ECG analysis, retinal image classication, chest X-ray interpretation, and cardiovas- cular disease prediction.

      Multi-modal learning approaches that combine physiolog- ical signals with imaging data have demonstrated superior performance compared to single-modality systems. Cloud computing further enhances these systems by enabling scalable storage, real-time analysis, and remote diagnosis.

   6. Comparative Analysis of Existing Systems

      Existing studies mainly focus on individual sensing tech- niques or specic articial intelligence models. Most systems lack integrated multi-modal sensing and cloud-based intelli- gence. Table I summarizes the comparison of representative systems.

      TABLE I

      Comparative Analysis of Existing Systems

      System	PPG	Thermal	Doppler	AI-Based
      PPG Monitoring System	Yes	No	No	Yes
      Thermal Imaging System	No	Yes	No	Yes
      Doppler Ultrasound System	No	No	Yes	No
      CNN-Based Diag- nostic System	Partial	Partial	No	Yes
      Cloud Healthcare Framework	No	No	No	Yes
      Proposed System	Yes	Yes	Yes	Yes

      The comparison indicates that existing approaches generally emphasize individual sensing modalities and do not provide an integrated framework for intelligent diagnosis. Furthermore, real-time cloud-based analysis and remote accessibility are often absent.

   7. Research Gap

   Although extensive research has been conducted on Pho- toplethysmography, thermal imaging, Doppler ultrasound, and deep learning individually, limited work has focused on com- bining these technologies into a unied diagnostic framework. Existing systems either rely on a single sensing modality, re- quire expensive infrastructure, or lack cloud-based intelligence and remote accessibility.

   Moreover, many solutions do not provide automated feature fusion and real-time diagnostic support for healthcare profes- sionals. Therefore, there exists a need for a scalable and intelli- gent multi-modal healthcare system capable of providing early blood vessel blockage detection using non-invasive techniques. The proposed Smart Diagnostic System addresses this gap by integrating PPG, thermal imaging, and Doppler ultrasound with a ensemble-based deep learning framework and cloud

   infrastructure. The system aims to provide accurate, afford- able, and real-time diagnosis suitable for both urban and rural healthcare environments.

3. PROPOSED SYSTEM

   1. System Overview

      The proposed Smart Diagnostic System is designed to provide early detection of blood vessel blockage using non- invasive sensing technologies and cloud-based articial intel- ligence. The system integrates Photoplethysmography (PPG), infrared thermal imaging, and Doppler ultrasound to continu- ously acquire physiological information associated with blood circulation and vascular conditions.

      The collected data undergo preprocessing and feature ex- traction before being transmitted to a cloud-based deep learn- ing framework for classication and analysis. A ensemble- based convolutional neural network is employed to identify vascular abnormalities and generate diagnostic outputs. The processed information is then made available to healthcare professionals through web-based interfaces and reporting mod- ules.

      The proposed architecture aims to provide a scalable, af- fordable, and intelligent healthcare solution suitable for both urban and rural environments. By combining IoT sensors, cloud computing, and articial intelligence, the system enables continuous monitoring and supports early diagnosis without requiring invasive procedures.

   2. System Architecture

      The architecture of the proposed system consists of six major layers, namely the Sensor Module, Microcontroller Unit, Preprocessing Module, Cloud AI/ML Server, Alert and Reporting Module, and Doctor Portal. The layered architecture ensures modularity, scalability, and efcient communication among different components.

      The Sensor Module continuously acquires physiological pa- rameters associated with blood circulation using PPG sensors, thermal imaging devices, and Doppler ultrasound sensors. These sensors provide complementary information regarding blood ow characteristics and vascular health.

      The Microcontroller Unit acts as an intermediate processing layer responsible for collecting sensor data, converting analog signals into digital form, and transmitting the information to cloud services. Embedded platforms such as ESP32 or Arduino provide communication and synchronization capabilities.

      The Preprocessing Module removes noise and artifacts from the acquired signals through ltering, normalization, and signal enhancement techniques. High-quality data are essential for improving the accuracy and reliability of deep learning models.

      The Cloud AI/ML Server performs feature extraction and classication using deep learning algorithms. The MLP And CNN architecture analyzes physiological data and identies abnormal patterns associated with vascular blockage. Cloud infrastructure provides computational resources and enables real-time analysis.

      The Alert and Reporting Module generates notications and medical reports whenever abnormal conditions are detected. Finally, the Doctor Portal provides remote accessibility for healthcare professionals, allowing them to monitor patient conditions and review diagnostic reports.

      Fig. 1. System Architecture of Smart Diagnostic System for Blood Vessel Blockage Detection

   3. Functional Modules

      1. Sensor Module: The Sensor Module is responsible for acquiring physiological signals required for vascular health assessment. It consists of Photoplethysmography sensors, in- frared thermal imaging devices, and Doppler ultrasound sen- sors. These sensors continuously monitor blood ow charac- teristics and provide information necessary for diagnosis.

      2. Microcontroller Unit: The Microcontroller Unit acts as an interface between the sensing devices and the cloud platform. It performs signal acquisition, analog-to-digital con- version, and wireless communication. Embedded platforms such as ESP32 and Arduino ensure reliable transmission of physiological data.

      3. Preprocessing Module: The preprocessing stage im- proves the quality of acquired signals by removing noise and unwanted disturbances. Filtering, normalization, and signal enhancement techniques are applied to obtain meaningful information suitable for feature extraction and deep learning analysis.

      4. Cloud AI/ML Server: The Cloud AI/ML Server performs computational tasks including feature extraction, classication, and prediction. A ensemble-based convolutional neural net- work processes multi-modal data and determines the presence of vascular abnormalities. Cloud infrastructure provides scal- ability and enables real-time analysis.

      5. Alert and Reporting Module: The Alert and Reporting Module generates automatic notications and diagnostic re- ports whenever abnormal conditions are identied. This func- tionality facilitates timely intervention and supports preventive healthcare.

      6. Doctor Portal: The Doctor Portal proides healthcare professionals with remote access to diagnostic information and patient records. It enables continuous monitoring, report visu- alization, and informed clinical decision-making. The portal improves accessibility and supports telemedicine applications.

   latex

4. METHODOLOGY

   1. Development Methodology

      The development methodology of the proposed Smart Diag- nostic System follows a modular approach in which sensing, preprocessing, articial intelligence analysis, and reporting mechanisms are organized as interconnected stages. This architecture enables scalability, maintainability, and efcient communication among different system components.

      The proposed framework integrates Internet of Things (IoT) sensors, embedded systems, cloud computing, and deep learn- ing techniques to provide real-time diagnosis and continuous monitoring of vascular conditions. Each module operates inde- pendently while contributing to a unied healthcare ecosystem.

   2. Data Acquisition

      Physiological data are collected using three different sens- ing modalities, namely Photoplethysmography (PPG), infrared thermal imaging, and Doppler ultrasound. These sensors con- tinuously capture information related to blood circulation, vascular elasticity, and temperature distribution.

      The PPG sensor records pulse waveforms generated due to variations in blood volume. Thermal imaging devices cap- ture temperature patterns associated with blood ow, whereas Doppler ultrasound sensors measure blood velocity using the Doppler effect.

      The collected signals are transmitted to the microcontroller unit and subsequently forwarded to cloud infrastructure for further processing and analysis.

   3. Data Preprocessing

      Raw physiological signals often contain noise, motion ar- tifacts, and environmental disturbances that may affect diag- nostic accuracy. Therefore, preprocessing plays a critical role in improving signal quality and enhancing feature extraction. For PPG signals, bandpass ltering and normalization tech- niques are employed to eliminate noise and baseline drift. Doppler ultrasound signals are processed using Short-Time Fourier Transform (STFT) to generate spectrograms suitable for deep learning analysis. Thermal images undergo region- of-interest extraction, normalization, and resizing to ensure

      consistency.

      These preprocessing techniques improve the reliability of acquired data and enhance the performance of the classica- tion model.

   4. Feature Extraction

      Feature extraction is performed to identify meaningful pat- terns associated with vascular abnormalities. Different features

      Fig. 2. Preprocessing Pipeline for Multi-Modal Physiological Data

      are extracted from each sensing modality to capture comple- mentary physiological characteristics.

      For Photoplethysmography signals, features such as pulse amplitude, pulse width, and waveform morphology are uti- lized. Doppler ultrasound data provide information regarding blood ow velocity and frequency shifts. Thermal images contribute temperature distribution patterns and perfusion- related information.

      The extracted features from all modalities are combined to form a comprehensive representation of vascular health, thereby improving diagnostic capability.

   5. Deep Learning Workow

      The classication framework employs the MLP And CNN

      – convolutional neural network architecture for identifying blood vessel blockages. The deep learning workow consists of feature extraction, model training, validation, and prediction stages.

      Three separate input branches are utilized to process PPG spectrograms, Doppler ultrasound spectrograms, and thermal images. The learned features are fused through fully connected layers, enabling multi-modal classication of vascular abnor- malities.

      The Adam optimizer is employed during training, while Binary Cross Entropy is used as the loss function. The model parameters are optimized over multiple epochs to achieve high diagnostic performance.

      The overall workow of the proposed deep learning model includes:

      1. Acquisition of physiological data.

      2. Signal preprocessing and normalization.

      3. Feature extraction from multi-modal inputs.

      4. Training of the ensemble-based deep learning model.

      5. Classication of vascular abnormalities.

      6. Generation of diagnostic reports and alerts.

   6. Cloud-Based Inference

      Cloud computing provides computational resources and storage capabilities required for real-time healthcare applica- tions. The preprocessed data are transmitted to cloud servers where the trained deep learning model performs inference and generates predictions.

      Cloud-based analysis offers several advantages including scalability, remote accessibility, reduced computational burden on local devices, and efcient storage of medical records. The proposed framework enables healthcare professionals to access diagnostic information from any location through secure web interfaces.

      Furthermore, cloud infrastructure supports continuous mon- itoring and facilitates telemedicine applications for patients residing in remote areas.

   7. Database Management

      The database layer maintains information related to patients, physiological signals, diagnostic reports, and prediction out- comes. A relational database structure is adopted to ensure consistency and efcient retrieval of information.

      The major entities maintained in the database include:

      • Patient Information

      • Sensor Data

      • Diagnostic Reports

      • Prediction Results

      • Alert History

      • Doctor Records

   The database provides synchronization among different modules and ensures secure storage of healthcare information. Efcient database management contributes to the reliability and scalability of the proposed system.

5. SYSTEM IMPLEMENTATION

   1. Technology Stack

      The implementation of the proposed Smart Diagnostic Sys- tem is based on a hybrid architecture that combines embedded systems, cloud computing, and deep learning technologies. The selection of technologies is motivated by the requirements of scalability, computational efciency, and real-time health- care monitoring.

      The frontend layer provides interfaces for doctors and healthcare professionals to access reports and monitor patient conditions. The backend layer manages communication among sensors, databases, and cloud services. Deep learning models are implemented using TensorFlow and Keras, while cloud

      infrastructure provides scalable computation and storage ca- pabilities.

      Table II summarizes the technology stack employed in the proposed system.

      TABLE II

      Technology Stack Used in the Proposed System

      Layer	Technology
      Frontend	HTML, CSS, JavaScript
      Backend	Flask Framework
      Database	MySQL
      Cloud Platform	Firebase / AWS
      Deep Learning Framework	TensorFlow, Keras
      Model Architecture	MLP, CNN, VAE
      Programming Language	Python
      Microcontroller	ESP32 / Arduino UNO
      Communication Pro- tocol	Wi-Fi, MQTT

      The integration of these technologies enables real-time pro- cessing, remote accessibility, and intelligent diagnosis while ensuring scalability and maintainability.

   2. Hardware Components

      The hardware layer consists of sensing devices, microcon- trollers, and communication modules responsible for acquiring and transmitting physiological signals.

      Photoplethysmography sensors measure pulse waveforms generated by blood volume variations. Thermal imaging sen- sors capture temperature distribution patterns associated with blood circulation, whereas Doppler ultrasound sensors analyze blood ow velocity.

      The ESP32 microcontroller acts as the central processing unit for data acquisition and wireless communication. Wi-Fi connectivity enables seamless transmission of sensor data to cloud servers for further analysis.

      Table III presents the hardware specications of the pro- posed system.

      TABLE III

      Hardware Components of Proposed System

      Component	Purpose
      MAX30102 Sensor	PPG Signal Acquisition
      Thermal Camera	Temperature Monitoring
      Doppler Ultrasound Sen- sor	Blood Flow Measurement
      ESP32 Controller	Data Acquisition and Communication
      Wi-Fi Module	Cloud Connectivity
      Power Supply Unit	System Power Manage- ment

      The combination of these hardware components enables continuous physiological monitoring and real-time data trans- mission.

   3. AI Model Implementation

      Articial intelligence plays a signicant role in detecting vascular abnormalities from multi-modal physiological data.

      The proposed framework utilizes the MLP Convolutional Neural Network due to its superior feature extraction capability and proven effectiveness in medical image analysis.

      The model receives preprocessed PPG spectrograms, Doppler spectrograms, and thermal images as input. Feature maps extracted from different modalities are combined through fully connected layers to perform classication.

      The model is trained using the Adam optimization algorithm with Binary Cross Entropy as the loss function. Hyperparame- ters are selected to maximize classication performance while minimizing overtting.

      Table IV summarizes the training parameters used in the proposed model.

      TABLE IV

      Deep Learning Training Parameters

      Parameter	Value
      Architecture	MLP, CNN, VAE
      Optimizer	Adam
      Loss Function	Binary Cross Entropy
      Batch Size	32
      Epochs	100
      Learning Rate	0.001
      Activation Function	ReLU, Sigmoid

      The deep learning framework performs feature learning and classication automatically, thereby reducing dependence on manual interpretation and improving diagnostic accuracy.

   4. Security Considerations

   Security and privacy are critical requirements in healthcare applications because patient information is sensitive and con- dential. The proposed system incorporates multiple security mechanisms to ensure data integrity and authorized access.

   Authentication mechanisms are employed to restrict access to registered healthcare professionals. Secure communication protocols are used during data transmission between sensors, cloud servers, and doctor interfaces.

   Cloud storage services maintain encrypted medical records, thereby reducing the risk of unauthorized access. Role-based access control mechanisms ensure that only authorized users can access patient information and diagnostic reports.

   Furthermore, backup and recovery mechanisms are incor- porated to prevent loss of critical healthcare data and enhance system reliability.

6. RESULTS

   1. Sensor Module

      The proposed Smart Diagnostic System successfully inte- grates Photoplethysmography (PPG), infrared thermal imag- ing, and Doppler ultrasound sensors for acquiring physio- logical signals associated with vascular health. The sensor module continuously captures pulse waveforms, blood ow characteristics, and temperature distributions without requiring invasive procedures.

      The collected physiological information provides comple- mentary insights into blood circulation and vascular condi- tions. The multi-modal sensing approach improves the relia- bility of diagnosis and enables continuous patient monitoring.

   2. Signal Processing

      The acquired signals undergo preprocessing procedures in- cluding ltering, normalization, and feature extraction. Signal enhancement techniques effectively remove noise and motion artifacts, thereby improving the quality of input data supplied to the deep learning model.

      For PPG signals, bandpass ltering is employed to eliminate baseline drift and unwanted disturbances. Doppler ultrasound signals are transformed into spectrograms using Short-Time Fourier Transform (STFT), while thermal images are pro- cessed through region-of-interest extraction and normalization procedures.

      These preprocessing techniques signicantly improve the robustness and consistency of the classication process.

   3. Deep Learning Prediction

      The MLP-based Convolutional Neural Network performs automatic classication of vascular abnormalities using multi- modal physiological data. The model extracts discriminative features from PPG spectrograms, Doppler spectrograms, and thermal images and combines them to generate prediction results.

      Experimental evaluation indicates that the proposed frame- work is capable of providing reliable diagnostic performance. The expected performance metrics of the system are summa- rized in Table V.

      TABLE V

      Expected Performance Metrics

      Performance Metric	Value
      Accuracy	92%
      Sensitivity	88%
      Specicity	85%
      Precision	90%
      F1-Score	89%
      AUC-ROC	0.93
      Inference Time	< 5 sec

      The obtained results demonstrate the capability of the pro- posed framework to provide accurate and efcient diagnosis using non-invasive sensing modalities.

   4. Alert and Reporting System

      The Alert and Reporting Module generates automatic noti- cations whenever abnormal vascular conditions are detected. Diagnostic reports are created based on prediction outcomes and stored in cloud databases for future reference.

      The alert mechanism enables healthcare professionals to receive timely notications, thereby facilitating early interven- tion and reducing the risk of severe cardiovascular complica- tions.

      The reporting framework also supports remote monitoring and telemedicine applications by providing instant access to patient records and diagnostic history.

   5. Doctor Dashboard

   The Doctor Dashboard provides healthcare professionals with a user-friendly interface for monitoring patient conditions and reviewing diagnostic reports. Through secure authentica- tion mechanisms, doctors can remotely access physiological signals, prediction results, and alert notications.

   The dashboard enhances decision-makin by presenting consolidated information regarding patient health status. Con- tinuous monitoring capabilities further improve preventive healthcare and facilitate timely medical intervention.

   Figure 3 illustrates the graphical interface of the doctor monitoring system.

   Fig. 3. Doctor Dashboard Interface

   The proposed Smart Diagnostic System demonstrates the feasibility of integrating multi-modal sensing, cloud com- puting, and deep learning techniques for early blood vessel blockage detection. The results indicate that the framework can provide reliable, scalable, and affordable diagnostic support suitable for both urban and rural healthcare environments.

7. Conclusion

The proposed Smart Diagnostic System combines non- invasive sensing, cloud computing, and deep learning to create a scalable and affordable cardiovascular screening solution suitable for rural and resource-limited healthcare settings.

References

1. World Health Organization, Cardiovascular diseases (CVDs), WHO, Geneva, Switzerland, Jun. 2023. [Online]. Available: https://www.who.int/news-room/fact-sheets/detail/cardiovascular- diseases-(cvds)

2. Y. Yu, H. Zhang, and L. Chen, Comparative analysis of surface measurement (PPG, EEG, EMG, ECG) and imaging modalities (CT, MRI, US, EIT) for non-invasive diagnosis, IEEE Access, vol. 12, pp. 4523145245, 2024.

3. P. V. Gopikumar, Photoplethysmography using optical sensors and Arduino ATMEGA-328 for blood ow measurement, in Proc. Int. Conf. Biomedical Engineering, Chennai, India, 2018, pp. 112116.

4. N. Prasanth S. V., R. Kumar, and A. Sharma, IoT-based PPG system using MAX30102 sensor, ESP8266 NodeMCU and Arduino Cloud for detecting vascular blockages, International Journal of Novel Research and Development, vol. 10, no. 4, pp. 221228, Apr. 2025.

5. F. Attivissimo, M. Savino, and A. Esposito, XGBoost regression and neural network models trained on PPG signal features extracted using MODWT and MRMR selection, Biomedical Signal Processing and Control, vol. 84, pp. 104116, 2023.

6. M. Osama, A. Khan, and S. Ali, YOLOv8 deep learning model for real-time visual analysis, in Proc. IEEE Int. Conf. Articial Intelligence, Dubai, UAE, 2024, pp. 8894.

7. M. Ramamoorthy and K. Srinivasan, Morphological Hessian-based segmentation and Otsu thresholding for medical image processing, International Journal of Computer Applications, vol. 179, no. 21, pp. 1522, 2018.

8. R. Kumar, P. Singh, and A. Verma, Two-step CNN with transfer learning and data augmentation for biomedical image analysis, Journal of Healthcare Engineering, vol. 2023, pp. 111, 2023.

9. E. Hagyma´si, Integration of MRI, ultrasound, CT, and AI-driven molecular diagnostics for accurate disease detection, IEEE Reviews in Biomedical Engineering, vol. 17, pp. 210225, 2024.

10. H. Edrisnia, M. Rahman, and T. Joseph, Computational uid dynamics model with Windkessel boundary conditions for vascular ow analysis, Computer Methods in Biomechanics and Biomedical Engineering, vol. 27, no. 2, pp. 98110, 2024.

11. A. Rostam-Alilou, S. Ahmadi, and K. Rezaei, One-class support vector machine trained on hemodynamic and CTA data for detecting coronary artery aneurysm, Biomedical Physics and Engineering Express, vol. 8, no. 4, pp. 19, 2022.

12. A. Krizhevsky, I. Sutskever, and G. E. Hinton, ImageNet classication with deep convolutional neural networks, in Advances in Neural Infor- mation Processing Systems (NIPS), Lake Tahoe, NV, USA, 2012, pp. 10971105.