Issuu

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

BRAIN TUMOR DETECTION USING IMAGE PROCESSING Samarjeet Sangram Shinde1, Ashitosh Shivaji Sankpal2, Siddhali Bhanudas Desai3 123Computer Science & Engineering, Sanjay Ghodawat University, Kolhapur, India.

------------------------------------------------------------------------------***-------------------------------------------------------------------------can make accurate identification and delineation of Abstract-Assessment of a brain tumor using Magnetic tumors difficult. Resonance Imaging (MRI) is a crucial application of medical image processing. Tumors may vary In conventional clinical practice, specialists generally considerably in their size, shape, location, visual inspect medical images manually to locate abnormal appearance, and internal structure, making their regions and evaluate their characteristics. Although identification and analysis challenging. Differences in expert interpretation remains essential, examining a scan quality, image noise, and indistinct tumor large number of scans can require considerable time boundaries can further complicate manual examination and effort. Differences in individual observations may and increase the workload of medical professionals. To also result in variations in how tumor boundaries are address these challenges, a web-based Brain Tumor identified. These challenges have encouraged the Detection using Image Processing system has been development of computational approaches capable of developed to support automated MRI analysis. The assisting medical professionals with repetitive imageproposed application applies image preprocessing analysis activities. techniques to enhance and standardize medical images before subsequent analysis. Automated segmentation is The Brain Tumor Detection using Image Processing then performed to identify and outline regions suspected project proposes a web-based platform that applies of containing tumors. A 3D U-Net architecture is automated image-processing and machine-learning employed for tumor segmentation, while feature techniques to medical images. The system incorporates extraction is used to obtain important characteristics several processing stages, including image from the segmented region. The resulting information is preprocessing, tumor segmentation, feature extraction, subsequently provided to a survival prediction module and predictive analysis. Information obtained from the for additional analytical assessment. The system segmented tumor is additionally supplied to a survival combines Python, OpenCV, NumPy, prediction module, allowing the system to provide TensorFlow/PyTorch, Scikit-learn, React.js, and Redux prognostic information alongside tumor identification. to support image processing, deep learning, prediction, and web-interface development. By integrating these The proposed platform integrates medical image technologies, the application minimizes repetitive acquisition, preprocessing, automated tumor manual analysis and provides a structured platform for delineation, feature analysis, survival estimation, and displaying segmentation and prediction results. Overall, result visualization into a single workflow. This the proposed solution combines medical image integrated structure is intended to reduce repetitive preprocessing, automated tumor segmentation, feature manual examination and provide users with organized analysis, and machine-learning-based survival analytical results. For the frontend, React.js and prediction within a unified web application designed to Redux are utilized, whereas Python, OpenCV, support MRI-based brain tumor assessment. NumPy, TensorFlow/PyTorch, and Scikit-learn support image processing and predictive-model Keywords: Brain Tumor Detection, MRI, Medical development. Image Processing, Tumor Segmentation, 3D U-Net, Feature Extraction, Survival Prediction, Machine The primary objective of the application is to make Learning, Deep Learning, Python, OpenCV, NumPy, brain tumor analysis more efficient and accessible TensorFlow, PyTorch, Scikit-learn, React.js, Redux, through computer-assisted techniques. By combining Web-Based Application, Medical Image Analysis. image-processing techniques with deep learning and machine learning, the system can offer I. INTRODUCTION computational support which may help healthcare professionals in the assessment of medical images and Brain tumor analysis through medical imaging presents planning the next steps of treatment activities. [1], [2], a significant challenge because abnormal brain regions [5]. can differ substantially in terms of size, shape, texture, visual appearance, and anatomical location. Additional difficulties may arise from image noise, variations in scan quality, and unclear boundaries between tumor tissue and surrounding healthy tissue. These factors

Impact Factor value: 8.315

ISO 9001:2008 Certified Journal

Page 173

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

A. Working of Brain Tumor Detection using Image Processing

deep learning architectures can be used to automatically detect and segment tumor areas within MRI scans.

The proposed application is based on a series of interconnected stages for analyzing brain images. The process starts with the upload of a medical image by the user using the web interface. The uploaded scan is subjected to some preprocessing to produce a better image quality and to reduce the variations that might affect the analysis.

The analysis could also be expanded beyond tumor localization. Characteristics collected from the segmented regions may be utilized for prognostic uses like survival estimation. Consequently, modern brain tumor analysis systems are integrating image segmentation with feature-based and predictive techniques.

The image is then subjected to preprocessing and sent to the automated segmentation module. This component identifies the possibly abnormal area and separates it from the rest of the brain tissue. Once the tumor region has been obtained, the system can be used to perform the feature extraction process, which is used to find out relevant features like size, shape, texture and location of the tumor.

Despite advancements in this field, the development of reliable automated systems still remains a significant challenge. Differences in imaging characteristics, the availability and quality of training data, and the model's computational resources can influence system performance. Therefore, appropriate testing and validation remain crucial in the development of computer-assisted medical imaging (CAMI) solutions. [1], [2], [5].

The extracted features are then provided to the survival prediction component. This module applies information pertinent to the tumor to estimate the outcome of survival. The resulting segmentation information, extracted features, and the results of the predictions are then displayed through the web interface.

1. Traditional Brain Tumor Analysis Techniques The traditional approach to the assessment of brain tumors is mostly based on human specialists using a computer system. Trained human experts use computer systems for the visual interpretation of medical images. During MRI examination, specialists check for abnormalities based on the following: the intensity of images, anatomical position, and the size and shape of the tumor. When necessary, the suspected tumor area may also be manually outlined to determine its extent for diagnosis and treatment planning.

The application is developed to deliver timely and organized outputs, reducing the amount of repetition of manual inspection. The major stages of the system, consisting of uploading the image, comprehensive preprocessing, tumor segmentation, feature extraction, survival prediction, authentication and interface responsiveness are taken into account during system testing. [1], [2], [5].

II.

Although manual examination provides the crucial advantage of professional clinical judgment, it becomes more demanding when a large number of medical scans need to be interpreted. Tumors with complex or poorly defined structures or boundaries can make manual delineation more challenging. In addition, there can be differences in the observations made by different observers due to variations in the boundaries identified for the same tumor.

LITERATURE REVIEW

Automated medical image analysis has become increasingly important in the assessment of brain tumors. In conventional MRI evaluation, specialists play a central role in examining scans, identifying abnormal structures, and assessing tumor-related characteristics. Although this approach benefits from clinical expertise, manually reviewing a large number of scans can be time-consuming and labor-intensive.

These restrictions have led to increased interest in automated image-analysis techniques. Computational approaches can help specialists by handling repetitive processing operations and providing structured representations of potentially abnormal regions. Such systems are intended to complement, not replace, professional medical interpretation. [1], [2].

To meet these requirements, researchers have investigated the following solutions: computational techniques with image preprocessing, tumor segmentation, feature extraction, and machinelearning-based prediction. The progress of deep learning has been further expanded. These methods are developed to facilitate models in learning complex patterns that are directly extracted from medical imaging data. Particularly,

Impact Factor value: 8.315

2. Automated Techniques for Tumor Segmentation Automated tumor segmentation is an important aspect of medical image analysis since it allows abnormal

ISO 9001:2008 Certified Journal

Page 174

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

tissue to be identified and isolated from surrounding healthy structures. Prior to segmentation, medical images can undergo several preprocessing operations to make them more convenient for computational analysis. This process plays an important role in computer-assisted brain tumor analysis and can support the identification of tumor regions for surgical planning.

the survival rate of patients that cannot necessarily be detected through the use of an MRI scanner. Information about the clinical history of the patient, the biology of the tumor, details regarding the treatment, as well as many other factors specific to the patient may also impact the results. That is why future developments of the system might incorporate other data too. [2], [3], [5].

Deep learning methods help in automated segmentation because they can recognize spatial and visual patterns from training data. For example, 3D UNet architectures are important in volumetric medical imaging since they learn in three dimensions and consider spatial relationships when analyzing MRI volumes.

A. Comparison and Analysis of Literature Review The reviewed approaches indicate a move from the manual analysis of brain tumors currently being used towards more advanced automated approaches in the analysis of medical images. Such conventional approaches have been very manual and computationintensive in carrying out routine processes. Deep learning techniques have been employed to further these approaches to ensure that automatic systems can identify complex patterns in imaging data. Analysis can now not only identify tumor location but also carry out feature extraction and survival predictions on data.

The proposed project uses an AI-based segmentation approach to automatically locate tumor areas and identify their boundaries. The use of automation can decrease the amount of time spent on this step, along with the manual delineation needed to generate a uniform depiction of the detected area. The project also takes into account several medical imaging modalities, such as MRI scans, within its broader system objectives.

Manual Brain Tumor Assessment vs. Automated Segmentation

However, the performance of the automatic segmentation depends on the quality of the data. There are a number of issues that could affect the quality of the data, such as the quality of the training data, the features of the tumors, imaging conditions, architecture of the model, and the evaluation process. All these factors could affect the performance of the segmentation. The tumors can be different and unique in their structure and appearance.

Brain tumors are manually identified through the evaluation of medical images by professionals that identify the area where abnormal tissue exists. Manual tumor identification involves much skill and may consume much time, particularly when several medical scans are being considered. Irregular tumors and areas that have unclear borders make the manual identification difficult. Automated segmentation approaches use computational models to identify and separate suspected tumor regions from the surrounding tissue. Such techniques can reduce repetitive manual work and generate structured segmentation outputs. However, automated models require appropriate training datasets and extensive validation before their results can be considered dependable for practical medical assistance.

3. Tumor Feature Extraction and Survival Estimation After segmentation of the tumor region, additional properties can be extracted based on the region identified. In the suggested framework, some of the features related to the tumor are size, shape, texture, and location. These features provide a more detailed representation of the detected tumor and can be used as input for subsequent analytical modules.

Accordingly, the automatic segmentation technique should be considered as a tool that provides assistance in addition to clinical interpretation, but not as a substitute for clinical skills. [1], [2].

Moreover, the obtained features are also used for survival prediction using machine learning. In contrast to the restriction in determining the position of anomalous tissue, the proposed approach uses the data associated with tumors to predict the outcome of patient survival. Thus, the use of the developed application is expanded not only to the task of image segmentation but also to the disease prognosis.

Image Processing Techniques vs. Deep Learning Methods Traditional methods in image processing usually employ processes like preprocessing, enhancement, segmentation, and feature extraction. This kind of

On the other hand, there are some parameters affecting

Impact Factor value: 8.315

ISO 9001:2008 Certified Journal

Page 175

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

process can be helpful in preparing and extracting valuable information from medical images. It is quite hard to come up with an effective image processing algorithm on one’s own. The process might not be effective enough to handle large datasets and diverse characteristics of the tumor.

knowledge by developing automatic methods for segmenting tumors and predicting survival of patients. The above features have been incorporated in a webbased application, whose entire process involves medical image acquisition, preprocessing, tumor segmentation, feature extraction, survival prediction, and presentation of the results in an interactive manner.

Deep learning offers an adaptable method, as models can be taught to find patterns within the image space, which are important for the task at hand based on training examples. The use of volumetric imaging, together with 3D U-Net, helps the algorithm to find the spatial information

The frontend is built with React.js and Redux, whereas Python, OpenCV, NumPy, TensorFlow/PyTorch, and Scikit-learn are used for image processing and prediction purposes respectively. With these technologies combined, the system sets up an organized.

contained within MRI volumes and perform segmentation of tumors. Whereas traditional image processing has beenfully utilized for the analysis, the approach under consideration has introduced the use of artificial intelligence and deep learning in the process of image analysis. The original images may be subjected to any form of image processing prior to automated analysis. The inputting of data may be done manually whereas deep learning is used for automated analysis. This is done through taking advantage of the pattern recognition from tumor segmentation training data [2], [4].

This combination of segmentation process and tumor features identification will help in reducing repetition and unnecessary work. Moreover, it will help in transferring the computed data to the healthcare professionals. On the other hand, the accessibility of the design of this system is improved through integration of all the processes in one software [1], [2], [5], [6].

III.

Tumor Segmentation vs. Survival Prediction

The design of this proposed system is an online brain tumor detection system where medical image analysis through the use of machine learning is done to help detect and analyze tumors. The general process starts with the BraTS dataset that has been preprocessed in preparation for the development of the model. The preprocessed data is used to train and segment the tumors using the 3D U-Net model.

Brain tumor segmentation and brain tumor survival prediction involve two distinct yet closely connected aspects of brain tumor analysis. Brain tumor segmentation is concerned with the localization and demarcation of the tumor, whereas brain tumor survival prediction deals with using the knowledge about the tumor to predict a probable patient outcome. The above features can be included in one software that will allow one to have structural knowledge about the tumor and prognosis. In the suggested framework, the features extracted from the segmented region will be given to the survival predictor part, thus giving extra functionality to the current software other than tumor diagnosis [2], [3].

After training the model, it can be used to analyze medical images to segment the areas that have been segmented as tumors. This allows the expansion of the system for tumor analysis through the extraction of features of the tumor, which is used as the input to the survival prediction component. This is how all the components of the pipeline come together.

A. Overall Analysis

This general pipeline encompasses the processes of preparing the datasets, computer-assisted model training, tumor segmentation, feature extraction, and survival prediction.

According to literature, there is a tendency towards automating the process of brain tumor analysis. The conventional examination method still remains relevant because of the necessity of having specialized knowledge. However, the manual method may be laborious, particularly in cases when multiple images should be analyzed.

The web application offers an interface through which users can upload and access their medical images for analysis purposes. The scans that have been uploaded are first subjected to preprocessing to make them ready for use in subsequent machine-learning algorithms. The automatic segmentation module then detects the part that seems to be affected by the tumor.

Image processing techniques serve as the basis for deriving information out of medical images. Machine learning/deep learning techniques further add to this

Impact Factor value: 8.315

PROPOSED SYSTEM

ISO 9001:2008 Certified Journal

Page 176

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

The information obtained about the tumor is used in the feature extraction and prediction modules to give survival results.

information and makes predictions about the outcome of patient survival. The proposed methodology not only detects the tumor but also exploits the segmented tumor information to perform additional clinical analysis and survival prognosis. The sequential process makes sure that the information gathered in the segmentation process is applied properly in the remaining processes of the proposed system.

Combining the above stages in one software, the suggested design offers a structured process of analyzing brain tumors using computers. These features are used for producing segmented images, and results of the output and prediction can be made available to users via the web-based interface. This will make the task of analyzing images less repetitive and computationally easier.

Fig -1: Proposed System The proposed algorithm starts with the step of obtaining the BraTS dataset, preprocessing, and model training, which are shown in Fig. 1 below. The preprocessed data is used to train the 3D U-Net model. Then, having been trained, the model performs the task of tumor segmentation and provides the image of the segmented tumor. Thus, the following operations are performed:

Fig -2: Flow Diagram

BraTS Dataset → Data Preprocessing → 3D U-Net Model Training → Tumor Segmentation → Segmented Image

The flow diagram of processing an MRI scan depicted in Fig. 2 below indicates the sequence of steps taken during the procedure starting from obtaining the image to the final result. The first step involves obtaining the MRI image that is then passed to the preprocessing step.

After that, the obtained segmented image data will be used to perform the feature extraction for survival prediction.

These are some of the procedures involved before the image gets processed to enhance its quality from the scanning. Some of the procedures include text removal, noise reduction, image smoothing, and image enhancement. This way, it becomes possible to remove any unwanted information or distraction that was present in the scanned image.

After segmentation of the tumor, the segmented tumor region is further processed to obtain important features from the MRI data. These features give information about the nature of the detected tumor and its size. The extracted features are then fed into the survival prediction model, which then processes the

Impact Factor value: 8.315

The enhanced image is then subjected to the

ISO 9001:2008 Certified Journal

Page 177

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

segmentation process in order to separate the tumor tissue from the rest of the brain. The morphology process is next applied to the segmented image in order to enhance the segmented region. The resultant information from the morphology process is then outputted as the final stage of the system.As such, the entire process can thus be described as follows:

segmentation data, feature extraction data, and survival prediction results. In addition, this module allows for the development and management of customized web services. The Healthcare Administration and Data Analysis/Research modules analyze the information produced. In addition, it enables real-time processing, meaning that the collected data can be processed and updated instantly while the system is operating. These elements together create an interconnected system in which image processing, tumor recognition, prediction, data management, and user interaction come together as an interrelated whole, which possesses the ability to function as an integrated system.

MRI Scan → Preprocessing → Text Removal → Noise Reduction and Image Smoothing → Image Enhancement → Segmentation → Morphological Operation → Output The above steps are particularly targeted at providing a definite pathway through which a medical image can be processed from acquisition to tumor region detection.

Fig -4: Use Case Diagram

Fig -3: Component Diagram

The above use case diagram depicts the interaction between the Brain Tumor Analysis System and the major users of the system in addition to other user groups.

The components are described by the component diagram, which also illustrates the functional units of the proposed brain tumor analysis system. The diagram shows how information is presented to the users and how the data is exchanged between the components.

Clinician The clinician primarily participates in the process of medical image analysis. This involves uploading the image, preprocessing of the image, and accessing the results of the automated tumor segmentation. Following the segmentation of the tumor, the feature extraction takes place, providing the survival prediction with the inputs. The analysis result is then presented to the clinician in a result-viewing interface. Healthcare Administrator.

In the User Interface, users are able to upload the medical images and view the results obtained from the system. First, the scan is sent to the Image Input and then to the Preprocessing module (after preprocessing is finished). The image coming out of the Preprocessing module is passed to the Automated Tumor Segmentation module. The segmentation output is then subjected to the Feature Extraction stage, wherein features of the tumor are extracted from the image. The extracted features are then provided to the Survival Prediction system, which then predicts the output.

The Healthcare Administrator job includes administrative tasks in the application. These procedures comprise setting up the application’s parameters and managing users. Generally, the Healthcare Administrator is responsible for administrative processes in the application.

The Database module stores information derived during the entire process workflow, including tumor

Impact Factor value: 8.315

ISO 9001:2008 Certified Journal

Page 178

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

Data Analyst/Researcher

development of the application.

The Data Analyst/Researcher can analyze the system’s generated information for further analysis and research. The use case functionality can be employed to analyze the processed results beyond the clinical workflow.

Consequently, the entire data processing can be represented by the following sequence: MRI Collection → Dataset Verification → Data Visualization → Preprocessing → Multichannel Image Creation → Training/Test Split → 3D U-Net Training → Validation → Prediction → UI Integration → Deployment

The overall use case architecture is associated with clinical, administrative, and research operations. The below-given diagram shows the relationship between image upload and preprocessing, segmentation, feature extraction, results visualization, and survival prediction with the system administration and research analysis functions in the proposed system.

The figure below represents the complete medical image processing workflow, beginning with data preparation, followed by model generation, prediction, interfaces, and implementation.

Fig -5: Data Flow Diagram The first step is the collection of the brain magnetic resonance (MRI) image dataset by the system. The second step involves the quality control process in which the system will determine MRI scans that have T1, T2, T2ce, and FLAIR images. In cases where some MRI scans do not have these images, the third step will involve obtaining the missing information.

Fig -6: Class Diagram The most relevant data objects of the brain tumor detection system are presented in the class diagram below.

The data obtained from the acquired images will then be subjected to analysis followed by the preprocessing step which will involve scaling. The various data obtained from the different modalities will then be combined to form a multimodal image with multiple channels. The combined data will then be converted into a set of numbers representing the image.

The associations between the objects are indicated as well. The User object contains information about users’ data and authentication. The User object interacts with the Image object and the System Setting object. The Image object stores the data about the medical images uploaded to the system, including the image ID, the user ID, image path data, and the date of uploading.

After that, it checks if the required resources like the training and mask folder are in the system. At this point, the data is divided into an 80:20 ratio to form the training and testing sets. The 3D U-Net is trained using the training set where the data generator provides the inputs needed for the model to be trained. When the training process is ended, the model is used to related to the user interface, where the predicted results are visualized. The final stage is the

Impact Factor value: 8.315

The Image is processed before being analyzed by the system. As a result of the image processing, a Segmented Tumor object is formed, which stores information about the segmentation, including the segmentation ID, the image ID, image path, and the date of segmentation.

ISO 9001:2008 Certified Journal

Page 179

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

Finally, the segmented tumor is associated with the ExtractedFeature class, which stores information extracted from the tumor, for instance, size, shape, and texture.

downstream activities can begin. Once an image is processed, it then goes to the Automated Tumor Segmentation Module (ATSM). This is the process in which the ATSM analyzes the image data and segments the tumor volume. Once done with the segmentation process, the segmented tumor image goes to the Feature Extraction Module.

In turn, the extracted data is connected to the SurvivalPrediction class that stores the survival prediction data, such as the prediction ID, feature ID, probability, and prediction date.

At the feature extraction stage, the relevant tumor features are identified, extracted and passed on to the Survival Prediction Module for the prediction process to take place. Once this is done, the results are then passed to the User Interface (UI) for the clinician to view the results. At the same time, the Real-Time Processing Module keeps track of the information being processed within the system.

The class diagram below represents the relationship between the identified classes and demonstrates the data flow within the application’s objects: User → Image → SegmentedTumor ExtractedFeature→ SurvivalPrediction

→

The data flow is relevant because it ensures that the application will store user credentials and process the image data to extract information about the tumor. In addition, the extracted data will be used to predict the survival probability that will eventually be stored by the application.

The above process can be described as: Clinician → User Interface → Image Input → Preprocessing → Tumor Segmentation → Feature Extraction → Survival Prediction → Results Interface

The classes proposed describe the information flow in the brain tumor detection system. The results of each step are fed into the next step, so an uploaded image can be carried through from the tumor segmentation and feature extraction to the prediction of survival. This will keep the link between the image and the processed image in place, and will simplify the addition of new image processing blocks or new features.

The system is designed so that at any given point in the procedure there is enough information generated by the previous step for the next step to commence. This way, there is a seamless flow of information and image data throughout the application.

IV.

RESULTS

This brain tumor detection and survival prediction system employs medical data, and the resulting model predicts the survival probability of a patient. The system performs image preprocessing, automatic tumor segmentation, tumor delineation, feature extraction, survival prediction, and web-based visualization. A medical imaging workflow was established, medical images were received and subjected to preprocessing procedures and automatic segmentation for the identification of abnormal regions. Further analysis can be performed to extract features such as size, shape, and texture from the segmented tumor data for survival prediction purposes.

Fig -7: Sequence Diagram The sequence diagram below shows the order of events and the communication of the components involved in the processing of medical images.

A web-based interface allows the uploading of medical images and provides access to the resulting data. A number of features were incorporated to serve healthcare clinicians, administrators, and researchers.

The process starts when the Clinician uploads an image through the User Interface. The uploaded image is received by the Image Input Component which passes the image to the Preprocessing Module where preprocessing activities are done before other

Impact Factor value: 8.315

The testing process considers the major functional components of the system, including:

ISO 9001:2008 Certified Journal

Page 180

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

Medical image uploading Image preprocessing Tumor segmentation Feature extraction Survival prediction User authentication Interface responsiveness Other input scenarios can also be considered for the test, including the upload of valid and invalid images, the presence or absence of tumors, the size and shape of tumors, and survival prediction. The modules’ tests aim to ensure the accuracy of each separate part’s output, as well as the accuracy of data transfer from one stage to another. Survival Prediction Performance The survival prediction component was evaluated using classification accuracy and median error. The model achieved the following results:

Evaluation Stage Classification Accuracy

Median Error

Validation

46.4%

217.92

Testing

61.0%

181.37

Fig -8: Ground-truth and predicted tumor segmentation results. Figure 8 represents the comparison of the visual results of the ground-truth tumor segments and the predictions of the proposed 3D U-Net model. The ground truth consists of the true tumor segments identified in the annotated images, while the prediction images reveal the sections detected by the 3D U-Net algorithm. The figure below consists of multiple MRI scans necessary to juxtapose the predicted and ground-truth tumor segments.

The accuracy of classification was 46.4% for the validation set and 61.0% for the testing set. The median error was 217.92 for validation and 181.37 for testing. Based on the findings, it is possible to conclude that the survival prediction tool under consideration was able to produce reliable results for the current data set. However, it should be noted that further studies need to be conducted before actual clinical use. For instance, the model’s accuracy can be improved if more data are used for the analysis.

The image above shows that the predicted segments possess similar configurations to the true segments. Therefore, one can conclude that the 3D U-Net was effective in predicting the tumor segments based on MRIs. The figure represents the qualitative result of the segmentation process since it demonstrates the relationship between the tumor segments (the true ones) and the predicted tumor segments.

On the other hand, the results of the current research should be regarded with caution for several reasons. First, it is evident that the power of the input data affected the outcome of the predictive model. In other words, the size of the data set influences the predictive performance, and a small sample size might negatively impact the final result. To elaborate, the current data set might lack sufficient data on tumors of different sizes and shapes, clinical findings, and imaging difficulties. As a result, there is a significant possibility that the new survival prediction tool will produce different results when used in other clinics with dissimilar data sets.

Impact Factor value: 8.315

Overall Results Analysis In general, the presented system is the proof of concept of the application of numerous medical image processing and analysis techniques with machine learning in one web-based application. In particular, the implementation of automatic image segmentation, feature extraction, survival prediction, and the corresponding visualizations could be identified as the basic set of tools for the computer-assisted brain tumor evaluation.

ISO 9001:2008 Certified Journal

Page 181

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

Indeed, the software under analysis was equipped with a tumor segmentation module that could identify tumor regions, while the feature extraction module implemented the description of these regions. In addition, the survival prediction module added the possibility to predict patient survival based on a set of extracted features.

improvement can include the expansion of the training set of the survival prediction model and the use of more advanced network architectures along with a thorough validation strategy. Future Scope Several improvements can be considered for future versions of the application.

At the same time, the analysis of performance demonstrated that the current level of predictiveness is not satisfactory. Therefore, it could be concluded that the way forward for the discussed system is the improvement of its performance through the expansion of the training set and the development of the machine learning algorithm.

1. Additional Imaging Modalities Firstly, additional imaging modalities, including DWI, perfusion-weighted MRI, PET, and MRSI, can be included in the system under consideration. Such an enhancement would make it possible to obtain more information about the observed tumors and improve the segmentation results.

To sum up, the created system could be considered as a basis for further improvement and development due to the utilization of modern image segmentation and feature extraction techniques, as well as the opportunity to predict patient survival. However, improving performance will require expanding the training set and enhancing the survival prediction module.

2. Advanced Deep Learning Architectures Secondly, it is necessary to explore more advanced deep learning architectures and structures, including attention-based, transformer-based, and multi-scale architectures. These architectures can improve the system's ability to analyze tumor characteristics and generalize more effectively.

V. CONCLUSION AND FUTURE SCOPE The Brain Tumor Detection and Survival Prediction system encompasses the complete process of computer-assisted brain tumor analysis using medical image processing and machine learning algorithms, which were trained on the BraTS database, using a 3D U-Net convolutional neural network architecture, applied to detect tumor regions in the given MRI images automatically.

3. Enhanced Survival Prediction Thirdly, the survival prediction module can be improved by introducing a combination of imaging features with other relevant factors (genomic, clinical, and treatment data). This would enable the system to provide more accurate predictions of patient prognosis.

After the tumor regions have been separated, the significant characteristic features that describe each of the cases can be extracted and used for further analysis by the survival prediction stage.

4. External Validation and Explainable AI Fourthly, to validate the results obtained, it is necessary to involve external data sets and apply XAI methods (attention maps, saliency maps, etc.). One of the ways to improve the system would be to make the decision-making process more transparent to increase clinicians’ trust in the program.

To ensure the effective use of the developed software, the whole system was implemented on a website, where users can upload their MRI scans and receive the results of the automatic preprocessing, segmentation, and feature extraction steps. In addition, the website offers multiple functions, including user authorization, medical image preprocessing, tumor detection, survival prediction model, data mining, and responsive design, among others.

5. Longitudinal Studies Finally, longitudinal studies should be carried out to track tumor progression over time to evaluate the system’s performance in longitudinal settings. 6. Real-Time Clinical Deployment

Although the developed application might be considered as an initial step in the automation of the entire process of brain tumor inspection, the survival prediction results highlight the need to improve and develop the project further. The key ideas of

Impact Factor value: 8.315

The proposed system can be further extended to a realtime clinical decision support system tool which can support radiologists and clinicians in the routine diagnosis process. Seamless connectivity with hospital

ISO 9001:2008 Certified Journal

Page 182

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 13 Issue: 08 | Aug 2026

p-ISSN: 2395-0072

www.irjet.net

information systems and medical imaging platforms can allow for automatic analysis of MRI scans, resulting in quicker turnaround times.

Intelligence and Data Communication: ICCIDC 2025, Springer, pp. 89–96.

BIOGRAPHIES

7. Model Optimization and Computational Efficiency

Samarjeet Sangram Shinde has completed his B.Tech in Computer Science and Engineering and has worked on various academic and technical projects in the field of computer science. His areas of interest include web development, artificial intelligence and software development. Currently, he is working as a Programmer Analyst Trainee at Cognizant.

Further optimization of the proposed model can be performed to reduce computational requirements and processing time. Techniques such as model pruning, quantization, and efficient network architectures could help deploy the system on systems with limited computational resources. 8. Larger and Diverse Datasets Larger, more diverse datasets with MRI scans of patients with varying clinical and demographic features can be added in future studies. Expansion of the training data set may lend greater confidence and transferability to the proposed system to different patient populations.

Ashitosh Shivaji Sankpal has completed his B.Tech in Computer Science and Engineering and has worked on various academic and technical projects in the field of computer science. His areas of interest include web development, artificial intelligence and software development. Currently, he is working as an intern in Python Full-Stack development.

REFERENCES [1] B. H. Menze et al., “The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS),” IEEE Transactions on Medical Imaging, vol. 34, no. 10, pp. 1993–2024, 2015, doi: 10.1109/TMI.2014.2377694.

Siddhali Bhanudas Desai has completed her B.Tech in Computer Science and Engineering and has worked on various academic and technical projects in the field of computer science. Her areas of interest include healthcare data analysis, claims processing, healthcare operations, and information technology. Currently, she is working as a Claim Analyst Trainee at Concentrix

[2] S. Bakas et al., “Identifying the Best Machine Learning Algorithms for Brain Tumor Segmentation, Progression Assessment, and Overall Survival Prediction in the BraTS Challenge,” arXiv preprint arXiv:1811.02629, 2018. [3] F. Wang, R. Jiang, L. Zheng, C. Meng, and B. Biswal, “3D U-Net Based Brain Tumor Segmentation and Survival Days Prediction,” in Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries, LNCS, vol. 11992, pp. 131–141, Springer, 2020, doi: 10.1007/978-3-030-46640-4_13. [4] A. Beers et al., “Sequential 3D U-Nets for Biologically-Informed Brain Tumor Segmentation,” arXiv preprint arXiv:1709.02967, 2017. [5] S. Bakas et al., “Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features,” Scientific Data, vol. 4, Art. no. 170117, 2017, doi: 10.1038/sdata.2017.117. [6] V. Thirthahalli Srinivasa et al., “Brain Tumor Detection and Segmentation,” in Proceedings of International Conference on Computational

Impact Factor value: 8.315

ISO 9001:2008 Certified Journal

Page 183