Model Klasifikasi Multikelas Kanker Serviks Menggunakan Citra Pap Smear Berbasis Interpretable Deep Learning

Annisa, Nurul (2026) Model Klasifikasi Multikelas Kanker Serviks Menggunakan Citra Pap Smear Berbasis Interpretable Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221031-Undergraduate_thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Kanker serviks masih menjadi tantangan kesehatan global yang serius, di mana skrining manual Pap smear masih terkendala oleh subjektivitas pengamat, keterbatasan waktu pemeriksaan, dan ketergantungan tinggi pada keahlian ahli sitopatologi. Sebagian besar sistem berbasis deep learning yang ada juga bersifat black box serta hanya mengandalkan citra sel tunggal tanpa konteks spasial dari citra patch aslinya. Penelitian ini bertujuan mengembangkan sistem diagnosis berbantuan komputer yang transparan, dan dapat diinterpretasikan untuk klasifikasi multikelas sel epitel serviks langsung dari citra patch Pap smear. Kontribusi utama penelitian ini terletak pada integrasi penuh tiga modul algoritmik dalam satu kerangka kerja end-to-end mulai dari deteksi Region of Interest (ROI), klasifikasi multikelas, dan dual-layer Explainable AI (XAI) yang menggabungkan interpretasi kualitatif sekaligus kuantitatif. Ide konseptual yang diusulkan adalah mempertahankan konteks spasial sel sejak deteksi ROI menggunakan YOLO11m, dilanjutkan pra-pemrosesan adaptif berbasis 5-Dimensional Spider Monkey Optimization (SMO), klasifikasi tiga kelas (Normal, Benign, Abnormal) menggunakan XceptionNet, serta validasi XAI melalui Grad-CAM++ dan segmentasi Adaptive K-Means untuk ekstraksi Rasio Nukleus-Sitoplasma (N/C Ratio). Hasil pengujian pada dataset SIPaKMeD menunjukkan model YOLO11m mencapai mAP50 sebesar 81,20%, enhancement adaptif meningkatkan SSIM hingga 0,9778, serta XceptionNet mencatatkan akurasi pengujian 98,45% dengan Precision dan Recall pada rentang 97-98%. Validasi XAI menghasilkan DSC segmentasi pada rentang 0,71-0,90, mengonfirmasi kesesuaian morfologis hasil model dengan anotasi ahli. Sistem ini berimplikasi sebagai kerangka kerja diagnostik yang dapat diintegrasikan dalam praktik klinis nyata melalui antarmuka GUI interaktif berbasis NiceGUI, mendukung transparansi dan kepercayaan klinisi terhadap penggunaan AI dalam sitologi kanker serviks.
====================================================================================================================================
Cervical cancer remains a serious global health challenge, where manual Pap smear screening continues to be hampered by observer subjectivity, time constraints, and heavy dependence on the expertise of cytopathologists. Most existing deep learning-based systems also operate as black boxes and rely solely on single-cell images, discarding the spatial context of the original patch. This study aims to develop a transparent and interpretable computer-aided diagnosis system for multiclass classification of cervical epithelial cells directly from Pap smear patch images. The primary contribution lies in the full integration of three algorithmic modules within a single end-to-end framework such as Region of Interest (ROI) detection, multiclass classification, and a dual-layer Explainable AI (XAI) module combining both qualitative and quantitative interpretation. The proposed conceptual design preserves spatial cell context from the ROI detection stage using YOLO11m, followed by adaptive pre-processing guided by 5-Dimensional Spider Monkey Optimization (SMO), three-class classification (Normal, Benign, and Abnormal) using XceptionNet, and XAI Validation through Grad-CAM++ and Adaptive K-Means segmentation for Nucleus-to-Cytoplasm (N/C) Ratio extraction. Evaluation on the SIPaKMeD dataset shows that the YOLO11m model achieved a mAP50 of 81.20%, adaptive enhancement improved SSIM up to 0.9778, and XceptionNet recorded a Testing accuracy of 98.45% with Precision and Recall in the 97-98% range. XAI Validation yielded segmentation DSC between 0.71 and 0.90, confirming morphological consistency between model predictions and expert annotations. This system serves as a diagnostic framework integrable into real clinical practice through an interactive NiceGUI-based interface, supporting transparency and clinician trust in the use of AI for cervical cancer cytology.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Explainable AI, Kanker Serviks, Pap Smear, SIPaKMeD, XceptionNet, YOLO11, Cervical Cancer, Explainable AI, Pap Smear, SIPaKMeD, XceptionNet, YOLO11
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RB Pathology T Technology > T Technology (General) > T58.62 Decision support systems T Technology > T Technology (General) > T59.7 Human-machine systems. T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Electrical Technology > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Nurul Annisa
Date Deposited:	31 Jul 2026 00:44
Last Modified:	31 Jul 2026 00:44
URI:	http://repository.its.ac.id/id/eprint/140301

Actions (login required)

View Item