Suryadi, Hulwah (2025) Pengembangan Sistem Prediksi Risiko Nodul Tiroid pada Citra Cine-clip Ultrasonografi Berbasis TI-RADS dan Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5023211034-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5023211034-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) |
Abstract
Kanker tiroid merupakan salah satu kanker dengan prevalensi tertinggi dalam tiga dekade terakhir. Adapun ultrasonografi (USG) umumnya digunakan sebagai modalitas utama dalam pemeriksaan nodul tiroid. Proses diagnosis yang masih sangat bergantung pada pengalaman dan jam terbang dari ahli radiologi menyebabkan proses diagnosis membutuhkan waktu cukup lama. Berbagai metode Computer-Aided Diagnosis (CAD) telah dikembangkan untuk membantu dokter spesialis radiologi dalam menganalisis citra USG nodul tiroid. Akan tetapi, sebagian besar dari penelitian sebelumnya hanya berfokus pada citra USG statis, dimana informasi yang disediakan kurang komprehensif karena hanya terpaku pada aspek spasial tanpa memperhatikan aspek temporal. Metode yang ada juga tidak dilengkapi dengan visualisasi, sehingga klasifikasi nodul dilakukan tanpa disertai penjelasan terkait fitur yang melatarbelakangi keputusan model. Hal tersebut dapat mengurangi kepercayaan dokter spesialis radiologi terhadap metode CAD karena tidak selaras dengan pendekatan proses diagnosis yang umum dilakukan. Dalam penelitian ini, telah dilakukan pengembangan model deep learning yang terintegrasi dengan panduan TI-RADS. Penelitian ini memanfaatkan citra sekuensial (cine-clip) USG dan informasi klinis yang memungkinkan model memahami karakteristik nodul secara lebih komprehensif. Klasifikasi citra sekuensial dilakukan dengan metode fine-tuned BiLSTM untuk memanfaatkan informasi temporalnya. Sementara itu, klasifikasi berbasis informasi klinis dilakukan dengan memanfaatkan Multilayer Perceptrons (MLP). Terakhir, model menentukan kategori nodul dan dilengkapi visualisasi terkait fitur TI-RADS yang melatarbelakangi keputusan tersebut. Penelitian ini mengembangkan metode prediksi risiko keganasan nodul tiroid yang terintegrasi karakteristik TI-RADS dengan tujuan meningkatkan akurasi dan efisiensi proses diagnosis serta kemampuan interpretasi model CAD. Metriks yang digunakan dalam proses evaluasi performa model meliputi confusion matrix, akurasi, presisi, sensitivitas, dan kurva ROC. Model menghasilkan akurasi sebesar 90% dengan mean of sensitivity and spesificity (MSS) senilai 75% pada data validasi. Sementara itu, pada data test model mencapai akurasi sebesar 74% dengan MSS senilai 0.62.
==================================================================================================================================
Thyroid cancer has become increasingly prevalent over the last 30 years, and ultrasonography (USG) is the primary tool for evaluating thyroid nodules. The diagnostic process is often lengthy and depends heavily on a clinician's expertise. Many Computer-Aided Diagnosis (CAD) methods have been developed to assist clinicians in analyzing Thyroid Nodule USG images. However, most of these systems use static images, which lack the temporal information found in sequential images (cine-clips). Additionally, existing methods often focus solely on binary classification (benign or malignant) without providing a visual breakdown of the TI-RADS features, which can reduce radiologists' confidence in the CAD method as it aligns differently from the radiologists' typical diagnosis process approach. In this study, a deep learning model integrated with TI-RADS guidelines was develope. Moreover, this research utilizes both thyroid nodule sequential (cine-clip) USG images and its clinical informations, which allow the model to understand the characteristics of the nodules more comprehensively. This sequential image computation uses the fine-tuned BiLSTM method to utilize its temporal information. Meanwhile, the clinical data computation uses Multilayer Perceptrons (MLP). The model will then determine whether the nodule is benign or malignant and provide visualizatio of the TI-RADS features behind the decision. This research develops a thyroid nodule malignancy risk prediction method integrated with TI-RADS characteristics to improve the accuracy and efficiency of the diagnosis process, as well as CAD model interpretability. Metrics used in the model performance evaluation process include confusion matrix, accuracy, precision, sensitivity, and ROC curve. This model achieved accuracy of 90% and mean of sensitivity and specificity (MSS) of 75% on the validation set and accuracy of 74% and MSS of 0.62 on the test set.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Computer-aided diagnosis (CAD), Nodul tiroid, TI-RADS, Ultrasonografi, Computer-aided diagnosis (CAD), Ultrasonography, Thyroid nodule, TI-RADS |
| Subjects: | R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine > RC78.7.U4 Ultrasonic imaging. |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis |
| Depositing User: | Hulwah Auliyah Suryadi |
| Date Deposited: | 08 Aug 2025 04:00 |
| Last Modified: | 08 Aug 2025 04:00 |
| URI: | http://repository.its.ac.id/id/eprint/126702 |
Available Versions of this Item
- Pengembangan Sistem Prediksi Risiko Nodul Tiroid pada Citra Cine-clip Ultrasonografi Berbasis TI-RADS dan Deep Learning. (deposited 08 Aug 2025 04:00) [Currently Displayed]
Actions (login required)
 |
View Item |