ANALISIS PERBANDINGAN CNN DENGAN COST-SENSITIVE LEARNING UNTUK DETEKSI TUBERKULOSIS PARU PADA CITRA X-RAY THORAX

Tuberculosis (TB) remains one of the heaviest infectious disease burdens in Indonesia, with an estimated 1.09 million cases and 125,000 deaths each year, while access to radiologists in primary health facilities is limited. This study compares two EfficientNetB0 models for classifying chest X-ray images as normal or TB: a baseline model trained without class weighting and a cost-sensitive model that applies class weights of 0.6 for the normal class and 3.0 for the TB class. The Tuberculosis (TB) Chest X-ray Database, containing 3,500 normal and 700 TB images, was divided into 3,000 training, 950 validation, and 250 test images (200 normal, 50 TB). On the test set, the baseline model achieved 97.20% accuracy, 98.00% sensitivity, 97.00% specificity, 89.09% precision, and a 93.33% F1-score, with 6 false positives and 1 false negative. The cost-sensitive model classified all 250 images correctly, reaching 100% on every metric with a ROC-AUC of 1.0000. The cost-sensitive model therefore performed better on this test set, mainly by eliminating false positives and the single missed TB case. The small test set, the single training run, and the absence of external validation limit these findings, so they need confirmation through repeated training with different random seeds, statistical testing, and external validation. ABSTRAK Tuberkulosis (TB) tergolong beban penyakit infeksi terberat di Indonesia dengan estimasi 1,09 juta kasus dan 125.000 kematian setiap tahun, sementara ketersediaan radiolog di fasilitas kesehatan primer masih terbatas. Penelitian ini membandingkan dua model EfficientNetB0 untuk mengklasifikasikan citra X-ray toraks sebagai normal atau TB, yaitu model baseline tanpa pembobotan kelas dan model cost-sensitive dengan bobot kelas 0,6 untuk kelas normal dan 3,0 untuk kelas TB. Tuberculosis (TB) Chest X-ray Database yang memuat 3.500 citra normal dan 700 citra TB dibagi menjadi 3.000 citra latih, 950 citra validasi, dan 250 citra uji (200 normal, 50 TB). Pada data uji, model baseline memperoleh akurasi 97,20%, sensitivitas 98,00%, spesifisitas 97,00%, presisi 89,09%, dan F1-score 93,33% dengan 6 false positive dan 1 false negative. Model cost-sensitive mengklasifikasikan seluruh 250 citra dengan benar sehingga seluruh metriknya 100% dengan ROC-AUC 1,0000. Dengan demikian, model cost-sensitive lebih unggul pada data uji ini, terutama karena tidak menghasilkan false positive dan tidak melewatkan kasus TB. Jumlah data uji yang kecil, pelatihan yang hanya dilakukan sekali, dan belum adanya validasi eksternal membatasi temuan ini, sehingga perlu dikonfirmasi melalui pelatihan ulang dengan beberapa random seed, uji statistik, dan validasi eksternal. Kata Kunci: citra X-ray toraks, convolutional neural network, cost-sensitive learning, EfficientNetB0, tuberculosis

Authors

Publication Details

Journal
eLEKTRIKA
Published
2026-10-30
DOI
https://doi.org/10.26623/elektrika.v18i2.16277
Primary Topic
COVID-19 diagnosis using AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

ANALISIS PERBANDINGAN CNN DENGAN COST-SENSITIVE LEARNING UNTUK DETEKSI TUBERKULOSIS PARU PADA CITRA X-RAY THORAX

Agus Supriyanto, Yusraka Dimas Al Iman, Septine Eka Putri
eLEKTRIKA
COVID-19 diagnosis using AI
article

ANALISIS PERBANDINGAN CNN DENGAN COST-SENSITIVE LEARNING UNTUK DETEKSI TUBERKULOSIS PARU PADA CITRA X-RAY THORAX

Agus Supriyanto, Yusraka Dimas Al Iman, Septine Eka Putri
article en

Abstract

Tuberculosis (TB) remains one of the heaviest infectious disease burdens in Indonesia, with an estimated 1.09 million cases and 125,000 deaths each year, while access to radiologists in primary health facilities is limited. This study compares two EfficientNetB0 models for classifying chest X-ray images as normal or TB: a baseline model trained without class weighting and a cost-sensitive model that applies class weights of 0.6 for the normal class and 3.0 for the TB class. The Tuberculosis (TB) Chest X-ray Database, containing 3,500 normal and 700 TB images, was divided into 3,000 training, 950 validation, and 250 test images (200 normal, 50 TB). On the test set, the baseline model achieved 97.20% accuracy, 98.00% sensitivity, 97.00% specificity, 89.09% precision, and a 93.33% F1-score, with 6 false positives and 1 false negative. The cost-sensitive model classified all 250 images correctly, reaching 100% on every metric with a ROC-AUC of 1.0000. The cost-sensitive model therefore performed better on this test set, mainly by eliminating false positives and the single missed TB case. The small test set, the single training run, and the absence of external validation limit these findings, so they need confirmation through repeated training with different random seeds, statistical testing, and external validation. ABSTRAK Tuberkulosis (TB) tergolong beban penyakit infeksi terberat di Indonesia dengan estimasi 1,09 juta kasus dan 125.000 kematian setiap tahun, sementara ketersediaan radiolog di fasilitas kesehatan primer masih terbatas. Penelitian ini membandingkan dua model EfficientNetB0 untuk mengklasifikasikan citra X-ray toraks sebagai normal atau TB, yaitu model baseline tanpa pembobotan kelas dan model cost-sensitive dengan bobot kelas 0,6 untuk kelas normal dan 3,0 untuk kelas TB. Tuberculosis (TB) Chest X-ray Database yang memuat 3.500 citra normal dan 700 citra TB dibagi menjadi 3.000 citra latih, 950 citra validasi, dan 250 citra uji (200 normal, 50 TB). Pada data uji, model baseline memperoleh akurasi 97,20%, sensitivitas 98,00%, spesifisitas 97,00%, presisi 89,09%, dan F1-score 93,33% dengan 6 false positive dan 1 false negative. Model cost-sensitive mengklasifikasikan seluruh 250 citra dengan benar sehingga seluruh metriknya 100% dengan ROC-AUC 1,0000. Dengan demikian, model cost-sensitive lebih unggul pada data uji ini, terutama karena tidak menghasilkan false positive dan tidak melewatkan kasus TB. Jumlah data uji yang kecil, pelatihan yang hanya dilakukan sekali, dan belum adanya validasi eksternal membatasi temuan ini, sehingga perlu dikonfirmasi melalui pelatihan ulang dengan beberapa random seed, uji statistik, dan validasi eksternal. Kata Kunci: citra X-ray toraks, convolutional neural network, cost-sensitive learning, EfficientNetB0, tuberculosis

eLEKTRIKAVol. 18(2)
Openalex Percentile: Top 13%
COVID-19 diagnosis using AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon