Non-Invasive Liver Fibrosis Detection from Ultrasound Images Using ConvNeXt V2 Tiny with CORAL-Based Ordinal Classification

Riyanto, Ananda Putra Wahyu (2026) Non-Invasive Liver Fibrosis Detection from Ultrasound Images Using ConvNeXt V2 Tiny with CORAL-Based Ordinal Classification. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) Cover.pdf Download (2MB)
	Text (Bab 1) BAB 1.pdf Download (246kB)
	Text (Bab 2) BAB 2.pdf Restricted to Repository staff only until 14 July 2028. Download (893kB)
	Text (Bab 3) BAB 3.pdf Restricted to Repository staff only until 14 July 2028. Download (796kB)
	Text (Bab 4) BAB 4.pdf Restricted to Repository staff only until 14 July 2028. Download (2MB)
	Text (Bab 5) BAB 5.pdf Download (228kB)
	Text (Daftar Pustaka) BIBLIOGRAPHY.pdf Download (275kB)
	Text (Lampiran) Lampiran.pdf Restricted to Repository staff only Download (597kB)

Abstract

Liver fibrosis is a progressive liver condition that requires accurate staging, while biopsy as the reference standard remains invasive and less suitable for repeated assessment. This study develops a non-invasive deep learning model for F0-F4 liver fibrosis staging from B-mode ultrasound images using ConvNeXt V2 Tiny with CORAL-based ordinal classification. The public “Liver Histopathology (Fibrosis) Ultrasound Images” dataset was audited and deduplicated using MD5 hashing to improve evaluation reliability. The proposed model uses ConvNeXt V2 Tiny as the backbone, with a CORAL ordinal head for F0–F4 staging and an additional binary head for advanced fibrosis risk (F≥ 3). Compared with the nominal softmax baseline, CORAL improved QWK from 0.7991 to 0.8133, reduced MAE from 0.4026 to 0.3680, and increased adjacent accuracy from 0.8788 to 0.9091. In the final holdout evaluation, the model achieved QWK = 0.8520, MAE = 0.3333, macro-F1 = 0.7584, adjacent accuracy = 0.9221, AUC = 0.9258, and sensitivity = 0.9565 for advanced fibrosis detection. Temperature scaling improved the Brier score from 0.1192 to 0.1174, while MC Dropout estimated prediction uncertainty. The model was implemented in the FibrosisRisk website prototype to present fibrosis stage, class probabilities, and advanced/non-advanced fibrosis status. Overall, this study shows that ConvNeXt V2 Tiny with CORAL ordinal classification provides a structured and interpretable framework for ultrasound-based liver fibrosis staging.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Muttaqin, Faisal	NIDN0030058602	faisalmuttaqin.if@upnjatim.ac.id
Thesis advisor	Puspaningrum, Eva Yulia	NIDN0005078908	evapuspaningrum.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.87 Neural computers

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Ananda Putra Wahyu Riyanto

Date Deposited:

15 Jul 2026 02:45

Last Modified:

15 Jul 2026 02:45

URI:

https://repository.upnjatim.ac.id/id/eprint/55447