Klasifikasi Stadium Fibrosis Hati pada Citra USG Menggunakan EfficientNet-B0
DOI:
https://doi.org/10.33005/santika.v6i1.1072Keywords:
fibrosis hati, ultrasonografi, deep learning, EfficientNet-B0, kalibrasi probabilitasAbstract
Penilaian stadium fibrosis hati penting dalam praktik klinis karena berhubungan dengan risiko komplikasi dan penentuan terapi. Namun, pembacaan citra ultrasonografi (USG) masih sangat dipengaruhi pengalaman operator. Penelitian ini menilai kemampuan EfficientNet-B0 dalam mengklasifikasikan lima stadium fibrosis hati (F0 - F4) menggunakan 6.323 citra USG. Data dibagi secara stratified menjadi data latih, validasi, dan uji dengan rasio 70:15:15. Sebelum pelatihan, seluruh citra diubah ke format RGB berukuran 224 × 224 piksel dan pada data latih diterapkan augmentasi berupa pembalikan horizontal serta rotasi kecil. Model dilatih dengan pendekatan transfer learning, fungsi loss berbobot kelas, optimizer AdamW, dan scheduler ReduceLROnPlateau. Penilaian kinerja dilakukan menggunakan accuracy, macro-F1, macro-AUROC, confusion matrix, serta evaluasi kalibrasi melalui reliability diagram dan Expected Calibration Error (ECE). Hasil pengujian menunjukkan bahwa model mencapai accuracy 0,9779, macro-F1 0,9671, dan macro-AUROC 0,9983, dengan 21 kesalahan klasifikasi dari 949 data uji. Model sangat baik dalam mengenali kelas ekstrem, tetapi masih mengalami kesulitan pada kelas menengah yang memiliki tampilan visual lebih mirip. Dari sisi probabilitas prediksi, nilai ECE 0,0161 menunjukkan bahwa tingkat keyakinan model cukup sesuai dengan akurasi aktual. Temuan ini menunjukkan bahwa EfficientNet-B0 memiliki potensi untuk digunakan sebagai dasar sistem bantu baca USG pada staging fibrosis hati, meskipun uji pada data eksternal tetap diperlukan sebelum penerapan klinis.
References
A. Berzigotti et al., “EASL Clinical Practice Guidelines on non-invasive tests for evaluation of liver disease severity and prognosis – 2021 update,” J. Hepatol., vol. 75, no. 3, pp. 659–689, Sep. 2021, doi: 10.1016/J.JHEP.2021.05.025.
A. J. Archer, K. J. Belfield, J. G. Orr, F. H. Gordon, and K. W. M. Abeysekera, “EASL clinical practice guidelines: non-invasive liver tests for evaluation of liver disease severity and prognosis,” Frontline Gastroenterol., vol. 13, no. 5, pp. 436–439, Sep. 2022, doi: 10.1136/flgastro-2021-102064.
Y. N. Zhang et al., “Liver Fibrosis Imaging: A clinical review of Ultrasound and Magnetic Resonance Elastography,” J. Magn. Reson. Imaging, vol. 51, no. 1, p. 25, Jan. 2019, doi: 10.1002/jmri.26716.
C. Fang and P. S. Sidhu, “Ultrasound-based liver elastography: current results and future perspectives,” Abdom. Radiol. (New York), vol. 45, no. 11, p. 3463, Nov. 2020, doi: 10.1007/s00261-020-02717-x.
S. K. Zhou et al., “A review of deep learning in medical imaging: Imaging traits, technology trends, case studies with progress highlights, and future promises,” Proc. IEEE. Inst. Electr. Electron. Eng., vol. 109, no. 5, p. 820, May 2021, doi: 10.1109/JPROC.2021.3054390.
K. Song, J. Feng, and D. Chen, “A survey on deep learning in medical ultrasound imaging,” Front. Phys., vol. 12, p. 1398393, Jul. 2024, doi: 10.3389/fphy.2024.1398393.
H. C. Park et al., “Automated classification of liver fibrosis stages using ultrasound imaging,” BMC Med. Imaging 2024 241, vol. 24, no. 1, pp. 36-, Feb. 2024, doi: 10.1186/s12880-024-01209-4.
Y. Joo, H. C. Park, O. J. Lee, C. Yoon, M. H. Choi, and C. Choi, “Classification of Liver Fibrosis From Heterogeneous Ultrasound Image,” IEEE Access, vol. 11, pp. 9920–9930, 2023, doi: 10.1109/ACCESS.2023.3240216.
M. Tan and Q. V Le, “EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,” May 24, 2019, PMLR. Accessed: Mar. 18, 2026. [Online]. Available: https://proceedings.mlr.press/v97/tan19a.html
M. Raghu, C. Zhang, J. Kleinberg, and S. Bengio, “Transfusion: Understanding Transfer Learning for Medical Imaging,” Adv. Neural Inf. Process. Syst., vol. 32, Oct. 2019, Accessed: Mar. 18, 2026. [Online]. Available: http://arxiv.org/abs/1902.07208
Y. Cui, M. Jia, T.-Y. Lin, Y. Song, and S. Belongie, “Class-Balanced Loss Based on Effective Number of Samples”.
M. Sokolova and G. Lapalme, “A systematic analysis of performance measures for classification tasks,” Inf. Process. Manag., vol. 45, no. 4, pp. 427–437, Jul. 2009, doi: 10.1016/j.ipm.2009.03.002.
C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger, “On Calibration of Modern Neural Networks,” Jul. 17, 2017, PMLR. Accessed: Mar. 18, 2026. [Online]. Available: https://proceedings.mlr.press/v70/guo17a.html
J. Nixon et al., “Measuring Calibration in Deep Learning,” Aug. 2020, Accessed: Mar. 18, 2026. [Online]. Available: http://arxiv.org/abs/1904.01685
