Klasifikasi Grade Tumor Ginjal Dari Citra CT Scan Menggunakan Convolutional Neural Network (CNN) Dengan Visualisasi Heatmap

Widyasetya, Naila Aqila (2026) Klasifikasi Grade Tumor Ginjal Dari Citra CT Scan Menggunakan Convolutional Neural Network (CNN) Dengan Visualisasi Heatmap. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221012-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Tumor ginjal merupakan salah satu jenis kanker yang memerlukan penanganan sesuai dengan tingkat grade-nya. Penentuan grade tumor umumnya dilakukan melalui pemeriksaan biopsi histopatologi yang hingga saat ini masih menjadi golden standard. Meskipun demikian, karena prosedur biopsi bersifat invasif, pendekatan non-invasif terus dikembangkan sebagai pelengkap dalam proses evaluasi klinis. Oleh karena itu, penelitian ini mengembangkan sistem klasifikasi grade tumor ginjal berbasis citra CT Scan dengan memanfaatkan deep learning sebagai alat bantu pendukung keputusan (decision support tool). Dataset yang digunakan berasal dari KiTS23 dan terdiri atas 366 kasus. Tahap pre-processing meliputi penerapan Hounsfield Unit (HU) Windowing pada rentang nilai [-79, 304], dilanjutkan dengan segmentasi multiclass menggunakan arsitektur U-Net untuk memisahkan area ginjal dan tumor. Hasil segmentasi digunakan untuk memperoleh Region of Interest (ROI) melalui proses cropping serta menghitung Centrality Index (C-Index) yang menggambarkan posisi relatif tumor terhadap pusat ginjal. Selain mendukung proses klasifikasi, C-Index juga bernilai klinis karena dapat menjadi pertimbangan dalam menentukan jenis operasi yang sesuai, yaitu pengangkatan sebagian atau seluruh ginjal, berdasarkan posisi tumor. Selanjutnya, citra ROI menjadi input bagi model ResNet-18 yang telah di-fine-tuning menggunakan strategi progressive unfreezing untuk mengklasifikasikan tumor ke dalam kategori low grade (grade 1 dan 2) dan high grade (grade 3 dan 4). Interpretabilitas model ditingkatkan melalui visualisasi heatmap menggunakan Grad-CAM pada lapisan konvolusional terakhir. Hasil pengujian menunjukkan bahwa model mencapai akurasi sebesar 94.55% dengan nilai AUC sebesar 0.9916. Visualisasi Grad-CAM memperlihatkan bahwa perhatian model secara konsisten terfokus pada area tumor, yang mengindikasikan bahwa keputusan klasifikasi didasarkan pada karakteristik yang relevan secara klinis. Dengan performa tersebut, sistem ini berpotensi menjadi alat bantu non-invasif yang mendukung klinisi dalam mengevaluasi grade tumor ginjal, dengan tetap menjadikan pemeriksaan histopatologi sebagai acuan utama.
====================================================================================================================================
Kidney cancer requires appropriate treatment based on its tumor grade. Tumor grading is generally determined through histopathological biopsy, which remains the current golden standard. However, because biopsy is an invasive procedure, non-invasive approaches continue to be developed as complementary tools to support clinical evaluation. Therefore, this study proposes a CT scan-based kidney tumor grading system using deep learning as a decision support tool. The dataset was obtained from KiTS23 and comprised 366 cases. The preprocessing stage included Hounsfield Unit (HU) windowing with a window range of [-79, 304], followed by multiclass segmentation using the U-Net architecture to separate the kidney and tumor regions. The segmentation results were then used to extract the Region of Interest (ROI) through a cropping process and to calculate the Centrality Index (C-Index), which represents the relative position of the tumor with respect to the center of the kidney. In addition to supporting the classification process, the C-Index also holds clinical value as it can serve as a consideration in determining the appropriate type of surgery, whether partial or total removal of the kidney, based on the tumor's position. The ROI images were subsequently fed into a fine-tuned ResNet-18 model employing a progressive unfreezing strategy to classify tumors into low-grade (grades 1 and 2) and high-grade (grades 3 and 4) categories. To improve model interpretability, Grad-CAM was applied to the final convolutional layer to generate heatmap visualizations. Experimental results demonstrated that the proposed model achieved an accuracy of 94.55% and an AUC of 0.9916. Furthermore, the Grad-CAM visualizations consistently highlighted the tumor region, indicating that the model based its classification decisions on clinically relevant features. These findings suggest that the proposed system has the potential to serve as a non-invasive decision support tool for assisting clinicians in kidney tumor grading while maintaining histopathological examination as the primary reference standard.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Tumor Ginjal, CT Scan, Convolutional Neural Network, ResNet, Grad-CAM; Kidney Tumor, CT Scan, Convolutional Neural Network, ResNet, Grad-CAM
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Naila Aqila Widyasetya
Date Deposited:	30 Jul 2026 00:55
Last Modified:	30 Jul 2026 00:55
URI:	http://repository.its.ac.id/id/eprint/140269

Actions (login required)

View Item