DETEKSI BATU GINJAL PADA CITRA MEDIS MENGGUNAKAN EVIDENTIAL NEURAL NETWORK DAN MEKANISME CONTEXTUAL CORRECTIONS BERDASARKAN KEPUTUSAN PARSIAL

Arga Bathara Dipoyono, NIM.: 22106050036 (2026) DETEKSI BATU GINJAL PADA CITRA MEDIS MENGGUNAKAN EVIDENTIAL NEURAL NETWORK DAN MEKANISME CONTEXTUAL CORRECTIONS BERDASARKAN KEPUTUSAN PARSIAL. Skripsi thesis, UIN SUNAN KALIJAGA YOGYAKARTA.

	Text (DETEKSI BATU GINJAL PADA CITRA MEDIS MENGGUNAKAN EVIDENTIAL NEURAL NETWORK DAN MEKANISME CONTEXTUAL CORRECTIONS BERDASARKAN KEPUTUSAN PARSIAL) 22106050036_BAB-I_IV-atau-V_DAFTAR-PUSTAKA.pdf - Published Version Download (5MB)
	Text (DETEKSI BATU GINJAL PADA CITRA MEDIS MENGGUNAKAN EVIDENTIAL NEURAL NETWORK DAN MEKANISME CONTEXTUAL CORRECTIONS BERDASARKAN KEPUTUSAN PARSIAL) 22106050036_BAB-II_sampai_SEBELUM-BAB-TERAKHIR.pdf - Published Version Restricted to Registered users only Download (11MB) | Request a copy

Abstract

Kidney stone detection using ultrasonography images is often hindered by variations in image quality and the subjectivity of interpretation. This study proposes the application of an Evidential Neural Network to classify renal ultrasonography images of patients with kidney stones and normal cases, by explicitly representing uncertainty through the Belief Functions theoretical framework. To improve the reliability of outputs produced by the Evidential Neural Network, this study models and applies a contextual correction mechanism using a partial decision approach, which works by grouping prediction outputs into specific decision areas and then applying appropriate correction parameters to each group. Model performance is evaluated using a Stratified K-Fold Cross Validation scheme with two specialized metrics, each measuring prediction error and utility-based predictive accuracy, respectively. Test results show that the model with a partial decision area-based correction mechanism achieves the best performance, with the lowest prediction error rate of 0.147 and the highest utility-based prediction accuracy of 0.91. The proposed approach significantly outperforms both the model without output correction and the model with correction applied without partial decision area grouping. These findings confirm that grouping predictions into partial decision areas is more effective in calibrating the model’s confidence level, resulting in more reliable diagnoses.

Share this knowledge with your friends :

Actions (login required)

View Item