Deteksi Dan Klasifikasi Kista Ovarium Pada Citra Ultrasound Menggunakan Efficientnetv2

A'yun, Kamala Qurrota (2026) Deteksi Dan Klasifikasi Kista Ovarium Pada Citra Ultrasound Menggunakan Efficientnetv2. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241114_Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Kelainan pada ovarium merupakan gangguan sistem reproduksi yang signifikan karena berdampak langsung terhadap tingkat kesuburan wanita. Mengingat posisi anatomisnya yang berdekatan dan saling berkaitan secara fungsional dengan rahim, pemeriksaan ovarium umumnya dilakukan secara simultan bersamaan dengan evaluasi rahim melalui metode ultrasonografi (ultrasound). Namun, interpretasi citra ultrasound ovarium menghadapi kendala teknis yang serius, terutama terkait kualitas citra yang sering kali mengandung speckle noise, kontras yang rendah, serta kemiripan visual antara struktur jaringan yang berbeda. Selain itu, diagnosis sangat bergantung pada kecakapan serta subjektivitas tenaga medis, yang meningkatkan risiko terjadinya variabilitas hasil interpretasi (inter-observer variability) dan potensi kesalahan diagnosis. Penelitian ini mengusulkan pengembangan sistem deteksi dan klasifikasi otomatis kelainan ovarium menggunakan arsitektur EfficientNetV2 untuk memberikan solusi diagnosis yang lebih objektif, konsisten, dan efisien. Meskipun EfficientNetV2 telah terbukti unggul dalam berbagai klasifikasi citra medis—seperti deteksi kanker payudara, tumor otak berbasis MRI, dan radiografi tulang—penerapannya pada deteksi kelainan ovarium masih memerlukan eksplorasi mendalam. Dalam penelitian ini, dataset yang digunakan terdiri dari 6.876 citra ultrasound ovarium yang dikategorikan ke dalam lima kelas: healthy ovary, dominant follicle, polycystic ovary, simple cyst, dan complex cyst. Untuk menjamin kualitas model, dilakukan proses data cleansing guna menghilangkan citra duplikat, sehingga meminimalkan potensi bias dan kebocoran data (data leakage). Dataset kemudian dibagi menjadi 70% data pelatihan, 10% data validasi, dan 20% data pengujian. Hasil eksperimen menunjukkan bahwa model EfficientNetV2 memiliki performa yang sangat baik dengan tingkat akurasi validasi mencapai sekitar 98%, disertai dengan nilai loss yang rendah dan stabil.
===================================================================================================================================
Ovarian abnormalities constitute a significant disorder of the female reproductive system because they directly affect fertility. Due to their anatomical proximity and functional relationship with the uterus, ovarian examinations are commonly performed simultaneously with uterine evaluations using ultrasonography (ultrasound). However, the interpretation of ovarian ultrasound images presents considerable technical challenges, particularly because image quality is often degraded by speckle noise, low contrast, and visual similarities among different tissue structures. Furthermore, diagnosis relies heavily on the expertise and subjective judgment of medical practitioners, increasing the risk of inter-observer variability and diagnostic errors. This study proposes the development of an automatic detection and classification system for ovarian abnormalities using the EfficientNetV2 architecture to provide a more objective, consistent, and efficient diagnostic solution. Although EfficientNetV2 has demonstrated superior performance in various medical image classification tasks, including breast cancer detection, brain tumor classification using MRI, and pediatric bone radiography, its application to ovarian abnormality detection remains underexplored. The dataset used in this study consists of 6,876 ovarian ultrasound images categorized into five classes: healthy ovary, dominant follicle, polycystic ovary, simple cyst, and complex cyst. To ensure data quality, a cleansing process was performed to remove duplicate images, thereby minimizing potential bias and data leakage. The dataset was subsequently divided into 70% training data, 10% validation data, and 20% testing data. Experimental results indicate that the EfficientNetV2 model achieved excellent performance, with validation accuracy reaching approximately 98%, accompanied by consistently low and stable loss values.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	EfficientNetV2, Ultrasound Ovarium, Klasifikasi, EfficientNetV2, Ovarium Ultrasound, Classification
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Kamala Qurrota A~yun
Date Deposited:	03 Aug 2026 07:41
Last Modified:	03 Aug 2026 07:41
URI:	http://repository.its.ac.id/id/eprint/139657

Actions (login required)

View Item