Cardio Advance: AI‑Powered Innovations for Angiogram Blockage

Detection System

Kavin Kumar D., Poovarasu K., Rajasekar J. and Santhosh Kumar K.

Department of Computer Science and Engineering Nandha Engineering College, Erode, India

Keywords: Angiogram Analysis, AI in Healthcare, Deep Learning, Arterial Blockage Detection, Convolutional Neural

Networks, Medical Image Processing, Cardiovascular Disease Diagnosis.

Abstract: Angiogram images serve a crucial role in the diagnosis of vascular diseases through blood x-ray and

identifying potential clogs. This project will provide a Computer Vision based automatic blockage detection

in Angiogram images using OpenCV and NumPy. Different picture preparing methods are incorporated with

the proposed technique, for example, grayscale transformation, brightness normalization, commotion

lessening (middle sifting), versatile thresholding, and morphological changes. The centre area instrument is

utilized in form extraction and Euclidean separate investigation, which discovers blood vessel shapes, detects

potential blockage areas by the vicinity. Which enhance brightness and variation to improve visibility, uses

adaptive thresholding to segment blood vessels and smooth the detected structures using dilation and

morphological operations. A final contour-based investigation determines possible occlusions Similarly, the

past uses the Euclidean distance among vascular edges. If the separate between the two forms is underneath

10 pixels, the framework will examine it as a potential blockage and feature it in the angiogram picture. Yield

gives an illustrated result with stamped intersections and argumentative message showing the only distance

or absence of snare. This robotized methodology provides a non-invasive, quick, and accurate strategy for

locating the blockage from an angiogram and helps therapeutic experts with an early determination and

treatment arranging.

1 INTRODUCTION

Among the leading causes of worldwide mortality is

cardiovascular maladies (CVDs), particularly

coronary supply route illness. Recognizing blockages

in supply routes at an early organize makes a

difference to intercede as before long as conceivable

to play down the chance of possibly life-threatening

ailments like heart assaults and strokes. The classical

strategy of analyzing an angiogram is regularly

subordinate on the translation of a restorative

proficient, making it both a time-consuming prepare

and one that's inclined to human blunder. To realize

this, we show CARDIO Development, an AI and

Computer Vision-based system for blockage

discovery in angiogram images in an mechanized

way. The framework employments progressed picture

preparing strategies like changing over to grayscale,

normalizing brightness, lessening clamor (middle

sifting), versatile thresholding and morphological

changes with OpenCV and NumPy. These strategies

move forward the perceivability of blood vessels,

empower exact blood vessel segmentation and

smooth edges of the vessels, subsequently permitting

exact blockage location. Euclidean remove

examination. In the event that the remove between two

vascular edges is less than a predefined edge, the

framework distinguishes this range as a plausible

blockage. A mechanized, non-invasive arrangement

that spares time, increases accuracy, and diminishes

human intercession. CARDIO Progress has the

potential to alter the scene of cardiovascular malady

location by getting to be an fundamentally portion of

clinical imaging conventions and workflows,

empowering healthcare suppliers to make prior

analyze and superior treatment choices, driving to

progressed understanding results amid the basic early

stages of cardiovascular malady.

2 RELATED WORKS

This think about investigates the application of fake

insights in restorative imaging for classification

546

D., K. K., K., P., J., R. and K., S. K.

Cardio Advance: AI-Powered Innovations for Angiogram Blockage Detection System.

DOI: 10.5220/0013932700004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 5, pages

546-552

ISBN: 978-989-758-777-1

division and conclusion progressing exactness and

productivity it highlights ai-based methods such as

profound learning and machine learning for robotized

infection detection1 this survey analyzes different

cardiovascular infection expectation models

emphasizing machine learning and measurable

approaches for hazard appraisal it compares show

exactness highlighting progressions and challenges in

prescient analytics for early diagnosis2 this ponder

investigates profound learning strategies for enrolling

demonstrative angiogram and fluoroscopy pictures

progressing arrangement precision the proposed

strategy improves image-guided mediations by

lessening enlistment mistakes and moving forward

visualization3 this think about proposes a profound

neural network-based approach for the mechanized

location of coronary course stenosis in x-ray

angiography the show improves demonstrative

exactness by proficiently recognizing stenotic

districts in angiographic images4 this work presents a

completely robotized framework leveraging neural

systems for translating coronary angiograms the show

progresses symptomatic accuracy by precisely

identifying and analyzing coronary course

abnormalities5 this consider presents a novel strategy

for extricating coronary supply routes and identifying

stenosis in obtrusive coronary angiograms the

approach upgrades symptomatic exactness by

moving forward supply route division and stenosis

identification6 this consider utilizes profound

learning-based protest discovery procedures for

robotized coronary supply route distinguishing proof

the approach improves exactness in identifying and

analyzing coronary course structures in therapeutic

imaging7 this think about centers on fragmenting

coronary supply routes from cat hub cuts utilizing

profound learning the proposed strategy makes

strides the exactness of supply route extraction for

way better symptomatic analysis8 this work presents

picture preparing calculations for identifying cardiac

blockages leveraging progressed methods for

progressed demonstrative precision the execution

improves computerized investigation in therapeutic

imaging9 this ponder centers on profound learning-

based division of the most vessel of the cleared out

front slipping fellow course in coronary angiograms

upgrading the precision of mechanized cardiac

diagnostics10 this think about presents a point-cloud-

based approach for mechanized 3d reproduction of

the coronary tree from x-ray angiography moving

forward visualization and examination of coronary

arteries11 this audit investigates robotized strategies

for recognizing myocardial ischemia and localized

necrosis centering on headways in machine learning

and picture handling techniques12 this consider

presents an mechanized symptomatic framework for

heart illness forecast utilizing manufactured neural

systems improving precision in early discovery and

diagnosis13 this consider investigates profound

learning procedures for coronary supply route

division in angiographic pictures moving forward

accuracy in restorative picture analysis14 this

consider centers on ai-based strategies for analyzing

coronary angiograms to identify stenosis improving

mechanized determination in cardiac imaging.

3 METHODOLOGY

3.1 DATA COLLECTION

CARDIO Development utilizes angiogram images

from medical databases, enhanced with GAN-

generated synthetic data to address data imbalance

and improve model generalization across diverse

imaging settings. Advanced preprocessing techniques

optimize diagnostic accuracy by reducing

computational complexity and enhancing contrast.

Grayscale conversion, brightness normalization, and

median filtering ensure optimal visibility and noise

reduction. Morphological operations such as dilation

and erosion refine vascular structures, enabling

precise deep learning-based blockage detection. This

structured approach enhances dataset optimization for

computer vision-based feature extraction and

analysis.

3.2 Feature Extraction using Computer

Vision Techniques

Following pre-processing, sophisticated image

processing with OpenCV and NumPy identifies

prominent vascular features from angiogram images.

Contour and edge detection improve visualization by

retaining fine blood vessel features, followed by

morphological enhancement to eliminate coarse

edges and noise reduction while maintaining critical

features. Adaptive thresholding automatically

distinguishes blood vessels in intricate or low-

contrast images, whereas gradient-based edge

detection emphasizes subtle changes in vessel width,

facilitating occlusion detection. A region-based

segmentation algorithm provides for correct stenotic

area extraction. This fast feature extraction pipeline

reduces false negatives and positives to provide

precise and automatic cardiovascular disease

diagnosis.

Cardio Advance: AI-Powered Innovations for Angiogram Blockage Detection System

547

3.3 Blockage Detection using Euclidean

Distance & Image Processing

Figure 1: Methodology.

CARDIO Advance employs state-of-the-art

image processing techniques combined with

Euclidean distance analysis to accurately measure

arterial narrowing and potential vascular

obstructions. The system evaluates vessel width by

measuring perpendicular distances across cross-

sections of the angiogram. The figure 1 shows the

methodology . If the Euclidean distance falls below a

predetermined threshold (e.g., 10 pixels), the region

is flagged as a potential blockage or stenosis.

Morphological operations such as dilation, erosion,

and closing refine vessel segmentation, reducing

noise and enhancing detection accuracy. The severity

of blockages is classified into four categories:

healthy, mild, moderate, and severe stenosis, aiding

in risk assessment and timely diagnosis. This method

offers a non-invasive, highly precise, and

computationally efficient approach for automatic

blockage detection, assisting doctors in early

diagnosis and treatment planning.

3.4 Classification using CNN-SVM

Hybrid Model

Cardio development employs a cnn-svm hybrid

model that unites the ability to learn deep features

with the high accuracy of conventional machine

learning methods vascular anomalies lumen wall

thickness and vascular shape are some of the spatial

and structural parameters that can be learnt

automatically by a convolutional neural network cnn

after recovery the fine features are passed through the

support vector machine svm which classifies

angiograms into four categories based on blockage

levels mild moderate severe and healthy this enables

the cnn to identify tiny variations between stenotic

and healthy regions because it learned pixel-level

variations and angiogram textures this integration

method improves classification accuracy eliminates

false positives and improves diagnostic consistency

3.5 Model Evaluation & Performance

Metrics

The execution examination of CARDIO Advance on

angiogram pictures assesses its adequacy in

classifying blocked courses. Measurements such as

exactness, exactness, review, and F1-score survey the

system's execution, guaranteeing negligible untrue

positives and wrong negatives. Exactness measures

accurately classified cases, whereas exactness

assesses the extent of genuine positives among all

anticipated positives. The table 1 shows Model

Evaluation & Performance Matrix. the Review

guarantees genuine positives are accurately

recognized, anticipating misclassification of ordinary

and blocked courses. The F1-score equalizations

exactness and review, improving blockage location

unwavering quality. Moreover, AUC-ROC evaluates

the model's capacity to recognize between solid and

deterred supply routes. The CNN-SVM half breed

demonstrate beats conventional strategies, making

strides classification precision and symptomatic

unwavering quality. This approach empowers a

speedier and more exact determination, helping

healthcare experts in early location and treatment

arranging.

Table 1: Model evaluation & performance matrix.

Model Accuracy (%) AUC-ROC Score

CNN 87.2% 0.90

sNet 89.1% 0.92

CNN+SVM 93.4% 0.96

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

548

3.6 System Architecture

The CARDIOADVANCE framework is designed as

a structured pipeline that ensures efficient and

accurate detection of arterial blockages in angiogram

images. The system operates through the following

key stages:

Figure 2: Proposed system architecture.

1. Data Acquisition & Augmentation – Angiogram

images are collected from medical databases, while

GANs generate synthetic images to address data

imbalance and enhance model diversity.

2. Preprocessing & Enhancement – Grayscale

conversion, brightness normalization, noise

reduction, and morphological operations refine vessel

structures for better segmentation and feature

extraction.

3. Feature Extraction – OpenCV-based techniques

detect vessel edges, extract contours, and apply

adaptive thresholding to highlight potential

blockages.

4. Blockage Detection – Euclidean distance analysis

quantifies arterial narrowing, with morphological

transformations refining detection and minimizing

artifacts.

5. CNN-SVM Classification: CNN extracts deep

vascular features, while SVM classifies blockages as

mild, moderate, severe, or healthy, enhancing

precision and reducing false positives.

6. Performance measurement: strong classification

is guaranteed through measures like accuracy,

precision, recall, F1-score, and AUC-ROC, that

assess system performance.

7. Integration into Medical Systems: By making

real-time angiography analysis possible, the model

assists doctors in planning treatments and making

early diagnoses.

By making CARDIO ADVANCE an extremely

accurate, non-invasive AI-based device, its design

revolutionizes the diagnosis of artery obstructions.

4 EXPERIMENTAL RESULTS

4.1 Experimental Dataset

The public angiogram repositories stacom coronary

artery angiography datasets and clinical datasets

supplied by research institutions are used for training

and evaluation of the proposed cardioadvance system

these datasets consist of high-resolution angiographic

images with different levels of arterial blockages

classified as having mild moderate and severe

stenosis motivated by the need for increasing model

generalization and avoiding overfitting we also

generate additional training data using generative

adversarial networks gans data augmentation

methods this process generates additional training

images artificially by applying transformations of the

following categories zooming randomly scale in or

out to simulate perspective in images change of

angles rotation flipping for make feature robust

contrast intensity modifications simulating changes

due to imaging conditions applying gaussian noise to

decrease reliance on highly specific patterns and

increase robustness.

Table 2, which parts the dataset into four sets,

appears that profound learning models can as it were

be profoundly exact and dependable with a expansive

and heterogeneous dataset boosted by generative ill-

disposed systems. Ordinary supply routes Gentle,

direct, and extreme stenoses Gans improves

demonstrative precision and empowers the

demonstrate to sum up over a assortment of

angiographic conditions by essentially boosting the

volume and differing qualities of information

Cardio Advance: AI-Powered Innovations for Angiogram Blockage Detection System

549

4.2 Results and Analysis

Standalone CNN and DL models are contrasted with

the proposed CNN-SVM hybrid model, which

utilizes Euclidean distance analysis and image

processing methods. The experimental results

indicate that CNN-SVM outperforms both standalone

CNN and conventional machine learning classifiers.

With 94.2%, CNN-SVM's accuracy was greater than

that of both standalone CNN (91.6%) and

conventional SVM (89.8%). Incorporating Euclidean

distance-based method enhances obstruction

classification by enhancing feature extraction. By

comparing the SVM classifier with traditional CNN-

softmax models, the latter decreases false positives

and false negatives while enhancing decision bounds.

Table 2: Dataset description.

Dataset Total Images Healthy Cases Blockage Detected Cases Severe Blockages

Public Angiogram 1,500 9,500 4,000 1,500

Clinical Dataset 7,000 4,500 2,000 500

nthetic Ima

es 5,000 2,500 2,000 500

GAN-Au

mente

8,000 4,000 3,000 1,000

Total 35,000 20,500 11,000 3,500

Figure 3: Accuracy vs epochs.

Figure 3: Accuracy vs Epochs The accuracy graph

in Figure 3 demonstrates that the proposed CNN-

SVM model achieves higher and more stable

accuracy over training epochs compared to

standalone CNN and traditional ML models.

CapsNet-enhanced feature extraction contributes to a

more progressive accuracy curve, reducing

fluctuations observed in conventional CNN training.

The Precision-Recall Curve (Figure 4) illustrates

the classification performance of the CNN-SVM

hybrid. The higher precision at varying recall levels

indicates superior classification capability, with

fewer false positives and false negatives than

standalone CNN and ML classifiers.

Figure 4: Precision-recall curve.

lattice displaying genuine positives untrue positives

genuine negatives and wrong negatives show

successfully minimizes wrong negatives making it

exceedingly reasonable for real-world angiogram

examination the heatmap visualization affirms that

the show accurately classifies a critical extent of

extreme blockage cases guaranteeing solid

cardiovascular conclusion certainty score

investigation the certainty score half breed show

makes strides at numerous stages preprocessing

increase upgrades picture differing qualities and

quality expanding certainty by 5-7 include extraction

with captures spatial progressions moving forward

certainty by 6-8 profound include extraction gives

improved representations boosting certainty by 8-10

choice boundary refinement assist improves

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

550

classification unwavering quality by 3-5 fine-tuning

with exchange learning regularization optimizes the

demonstrate encourage expanding certainty by 2-4

through these optimizations the cardio advance show

advances from an introductory 78-82 certainty after

crude information handling to a last optimized run of

92-95 guaranteeing profoundly solid blockage

discovery in angiograms.

Table 3: Performance analysis of ML and DL.

Classifie

Precision

(

)

Recall

(

)

F1-Score

(

)

Accurac

(

)

Decision Tree 70.00 69.00 59.00 67.00

Random Forest 71.00 71.00 65.00 71.35

Gradient Boosted Trees 68.00 73.00 70.00 73.44

CNN 91.07 87.68 89.32 88.83

CapsNet 93.25 89.43 91.30 90.15

CNN+SVM 95.12 92.36 93.72 91.50

Figure 5: Confusion matrix.

Figure 5 disarray lattice, presents the disarray.

5 CONCLUSIONS

The most recent AI system CARDIO Progress has

been implemented, with the point of revolutionizing

mechanized angiogram blockage discovery in terms

of symptomatic precision, preparing speed and

clinical unwavering quality. The application of cycles

of computer vision calculations, profound learning

models & Euclidean separate estimations permit the

framework to precisely identify, analyze and classify

the sort of blockages within the supply routes with

small to no help from a human. The real-time

preparing of large-scale angiogram information in

this way permits for early location, superior

conclusion and optimized treatment arranging, all of

which diminish the hazard of extreme cardiovascular

occasions. Assist advancements will be made in 3D

angiogram examination, prescient analytics, and

cloud-based arrangement for the system, expanding

its capabilities and empowering it to move well into

numerous diverse healthcare settings. The ceaseless

checking of patients, personalized chance evaluation

and information driven choice back will become

available due to consistent integration with electronic

wellbeing records (EHRs). The table 2 shows the

Table 3: Performance Analysis of Ml and Dl.

CARDIO ADVANCE is a progressive AI-powered

diagnostics arrangement for cardiovascular

examination to alter diagnostics scene all inclusive by

empowering an greatly productive, exact, and

operator-friendly, non-invasive investigation to

increase clinical workflows, maximize understanding

results, and maximize crisis reaction techniques in

cutting edge healthcare.

6 FUTURE WORK

CARDIOADVANCE will enhance blockage

detection using OpenCV and NumPy, refining

Euclidean distance-based analysis for improved

vascular stenosis estimation. Real-time processing

and cloud integration will enable remote AI-driven

diagnostics.Advanced feature extraction and adaptive

thresholding will ensure accurate and robust detection

across varying image qualities. A user-friendly web

and mobile interface will allow clinicians to upload

angiograms and receive AI assessments.Clinical

validation will enhance reliability, while optimization

efforts will reduce computational costs and

processing time. Explainable AI techniques will

ensure transparent and interpretable diagnostic

insights for healthcare professionals.

Cardio Advance: AI-Powered Innovations for Angiogram Blockage Detection System

551

REFERENCES

A Review of Cardiovascular Disease Prediction Models,

Vol. 16, February 2023, pp. 45-60.

A New Approach to Extracting Coronary Arteries and

Detecting Stenosis in Invasive Coronary Angiograms,

Vol. 70, No. 6, June 2023, pp. 1890-1901.

A Review of Automated Methods for Detection of

Myocardial Ischemia and Infarction, Vol. 16,Novembe

r 2023, pp. 789-802.

AI-Based Analysis of Coronary Angiograms for Stenosis

Detection, Vol. 28, No. 1, January 2024, pp. 123-134.

An Automated Diagnostic System for Heart Disease

Prediction using Artificial Neural Networks, Vol. 11,

December 2023, pp. 67890-67900.

Application of Artificial Intelligence for Classification,

Segmentation, and Diagnosis in Medical Imaging, Vol.

42, No. 1, January 2023, pp. 123-135.

Automated Detection of Coronary Artery Stenosis in X-ray

Angiography using Deep Neural Networks, Vol. 11,

April 2023, pp. 4567-4578.

CathAI: Fully Automated Interpretation of Coronary

Angiograms Using Neural Networks, Vol. 27, No. 5,

May 2023, pp. 2345-2356.

Deep Learning Approaches for Coronary ArterySegmentat

ion in Angiographic Images, Vol. 42, No. 12,Decembe

r 2023, pp. 4567-4578.

Deep LearningBased Registration of DiagnosticAngiogra

m and Fluoroscopy Images, Vol. 42, No. 3, March

2023, pp. 789-798.

Design and Implementation of Image ProcessingAlgorith

ms for Cardiac Blockage Detection*IEEE Access*,

Vol. 11, August 2023, pp. 56789-56799.

Machine Learning Techniques for Cardiovascular Disease

Diagnosis from Angiographic Data, Vol. 12, February

2024, pp. 2345-2356.

Object Detection for Automated Coronary Artery Using

Deep Learning, Vol. 11, December 2023, pp. 123456-

123467.

Point-Cloud Method for Automated 3D Coronary Tree

Reconstruction from Xray Angiography*IEEETransac

tions on Medical Imaging*, Vol. 42, No. 10, October

2023, pp. 3456-3467.

Segmentation of Coronary Arteries from CTA Axial Slices

using Deep Learning, Vol. 42, No. 7, July 2023, pp.

1789-1799.

Segmentation of the Main Vessel of the Left Anterior

Descending Artery in Coronary Angiograms using

Deep Learning, Vol. 70, No. 9, September 2023, pp.

2345-2355.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

552