Klasifikasi dan Estimasi Tingkat Keparahan Stroke pada Citra MRI Otak Menggunakan DenseNet121 dengan Integrasi Convolutional Block Attention Module (CBAM)

Putri, Safira Rahmalia (2026) Klasifikasi dan Estimasi Tingkat Keparahan Stroke pada Citra MRI Otak Menggunakan DenseNet121 dengan Integrasi Convolutional Block Attention Module (CBAM). Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (cover) 22083010073.-cover.pdf Download (1MB)
	Text (bab 1) 22083010073.-bab1.pdf Download (234kB)
	Text (bab 2) 22083010073.-bab2.pdf Restricted to Repository staff only until 15 September 2028. Download (763kB)
	Text (bab 3) 22083010073.-bab3.pdf Restricted to Repository staff only until 15 September 2028. Download (746kB)
	Text (bab 4) 22083010073.-bab4.pdf Restricted to Repository staff only until 15 September 2028. Download (1MB)
	Text (bab 5) 22083010073.-bab5.pdf Download (207kB)
	Text (daftar pustaka) 22083010073.daftarpustaka.pdf Download (204kB)
	Text (lampiran) 22083010073.lampiran.pdf Restricted to Repository staff only until 15 September 2028. Download (378kB)

Abstract

Stroke is one of the leading causes of death and long-term disability, requiring rapid and accurate detection and treatment. Magnetic Resonance Imaging (MRI) can be used to identify stroke lesions, but manual image analysis is time-consuming and can be subject to bias. This study aims to develop an automated pipeline for classifying and estimating stroke severity in MRI images using DenseNet121 with an integrated Convolutional Block Attention Module (CBAM), lesion segmentation using NVAutoNet, and integrating the analysis results into a Streamlit-based application. The ISLES 2022 dataset was processed through preprocessing stages and organized into four schemes based on combinations of DWI or RGB Fusion modalities and the “All Slices” or “Important Slices” strategies. The evaluation results show that the M3 scheme (DWI Important Slices) is the selected model, with an accuracy of 84.65%, a precision of 83.53%, a recall of 88.57%, an F1-score of 0.8598, and an AUC-ROC of 0.9195. Next, NVAutoNet was used to segment the images identified as positive, yielding a Mean Dice Score of 0.8347, Mean Precision of 0.8699, and Mean Recall of 0.8221 on 50 test data points. External validation on hospital MRI data showed that the M3 model successfully classified stroke and normal patients in accordance with radiological diagnoses using a slice probability threshold of 0.80 and a 50% majority rule. In the segmentation validation, the lesion mask results for stroke patients yielded a lesion volume estimate of 6.67 mL, which falls into the mild severity category. All classification, segmentation, volume estimation, and severity level results were then integrated into a Streamlit application to present the analysis results in an integrated manner.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Diyasa, I Gede Susrama Mas	NIDN197006192021211009	igsusrama.if@upnjatim.ac.id
Thesis advisor	Adziima, Andri Fauzan	NUPTK9844773674130292	andri.fauzan.fasilkom@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Q Science > QA Mathematics > QA76.6 Computer Programming
Q Science > QA Mathematics > QA76.87 Neural computers
T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Data Science

Depositing User:

Safira Rahmalia Putri

Date Deposited:

16 Sep 2026 06:26

Last Modified:

16 Sep 2026 08:52

URI:

https://repository.upnjatim.ac.id/id/eprint/60170