KLASIFIKASI SEL-SEL DARAH BERDASARKAN CITRA MIKROKOPIS MENGGUNAKAN CONVOLUTION NEURAL NETWORK SEBAGAI DASAR DIAGNOSA PENYAKIT ALL DAN AML OLEH DOKTER AHLI

CHARMA NOVIANTI H, Ika Candradewi, S. Si., M. Cs.

2022 | Skripsi | S1 ELEKTRONIKA DAN INSTRUMENTASI

Leukemia yang lebih dikenal dengan kanker darah termasuk salah satu kanker ganas. Leukemia berdasarkan maturitas sel immatur (blast) dibagi menjadi dua yaitu Acute Myeloid Leukaemia (AML) dan Acute Lymphoblastic Leukaemia (ALL). Proses deteksi penyakit tersebut memerlukan perhitungan sel-sel darah yang membutuhkan waktu dan keahlian tim medis. Penelitian-penelitian sebelumnya belum melakukan deteksi dan perhitungan sel limfoblast dan sel myeoblast beserta sel- sel darah lainnya seperti sel darah merah, sel darah putih normal dan platelet dalam melakukan deteksi penyakit ALL dan AML sehingga untuk analisa dan diagnosa penyakit lebih lanjut oleh dokter ahli kurang efektif. Penelitian ini bertujuan untuk klasifikasi sel-sel darah serta sel penanda penyakit ALL dan AML dalam satu pemodelan sistem berdasarkan citra mikrokopis apusan sel darah menggunakan metode convolution neural network (CNN) YOLO. Sistem deteksi yang terdiri dari 9 kelas sel darah ini menggunakan dataset 2.336 gambar mikrokopis apusan darah. Hasil evaluasi sistem terhadap data uji menggunakan model terbaik YOLOv5 menghasilkan performa sistem dengan akurasi 55,61%, presisi 58,94%, recall 90,1%, F1-score 71,27%, mAP 85,1% serta kecepatan komputasi 19 milidetik per-frame sedangkan ketika menggunakan model YOLOv4 sistem memiliki performa presisi 56%, recall 57%, F1-score 56%, mAP sebesar 31,49%, serta kecepatan komputasi sebesar 51 milidetik per-frame. Dengan menggunakan YOLOv5, sistem dapat mendeteksi sel penanda penyakit ALL dan AML dengan baik yaitu sel limfoblast sebagai penanda penyakit ALL memiliki akurasi 95,03%, presisi 86,33%, recall 83,3%, F1-score 84,77% dan mAP 92,9% sedangkan sel myeoblast sebagai penanda penyakit AML memiliki akurasi 97,23%, presisi 94,3%, recall 100%, F1-score 97,07% dan mAP 97,2%.

Leukaemia as known as blood cancer is one of the most malignant cancers. Leukemia based on the maturity of immature cells (blasts) is divided into two, they are Acute Myeloid Leukaemia (AML) and Acute Lymphoblastic Leukaemia (ALL). The process of detecting the disease requires calculation of blood cells which requires time and expertise of the medical team. Previous researches have not detected and counted lymphoblast and myeoblast cells with other blood cells such as red blood cells, normal white blood cells, and platelets in detecting ALL and AML diseases, so further analysis and diagnosis of ALL and AML diseases by doctors are less effective. The aims of the research is to classify blood cells and marker cells of ALL and AML cancers in a system modeling based on microscopic blood cell smear images using YOLO convolution neural network (CNN) method. Detection system that consists of 9 classess of blood cells used 2.336 smear blood microscopic images. System evaluation result by using the test data, the best model of YOLOv5 were able to get accuracy rate of 55,61%, precision rate of 58,94%, recall of 90,1%, F1-score of 71,27%, mAP value of 85,1% and with the computational speed rate of 19 miliseconds per-frame therefore for YOLOv4 were able to reach the precision rate of 56%, recall of 67%, F1-Score of 56%, mAP value of 31,49%, and with the computational speed rate of 51 miliseconds per-frame. By using YOLOv5, the system can detect ALL and AML diseases marker cells well. Lymphoblast cell as a marker of ALL disease were able to reach the accuracy rate of 95,03%, precision rate of 86,33%, recall of 83,33%, F1-score of 84,77%, mAP value of 92,9% while myeoblast cell as marker of AML diseases were able to reach the accuracy rate of 97,23%, precision rate of 94,3%, recall of 100%, F1-score of 97,07%, mAP value of 97,2%.

Kata Kunci : Leukemia, CNN, YOLO