Asisten Kanker Payudara

Portal Penelitian Breast Cancer

Abstrak

Kanker payudara adalah kondisi di mana sel-sel berkembang biak secara abnormal dan menyebar dengan sangat cepat ke bagian lain tubuh, terkadang bahkan menyebabkan kegagalan organ tertentu. Deteksi kanker payudara merupakan kondisi medis yang serius. Sangat penting untuk mendapatkan diagnosis dini agar dapat mengobati kondisi tersebut dan mencegah perkembangannya. Dalam tinjauan sistematis ini, kami meninjau total 53 studi untuk menyajikan tinjauan komprehensif tentang deteksi kanker payudara menggunakan deep learning, baik dengan metode machine learning klasik, deep learning, maupun metode ensemble. Penelitian ini melakukan tinjauan literatur sistematis (SLR) menggunakan paradigma PRISMA. Dari tahun 2019 hingga 2024, kami mencari database Scopus dan sumber terindeks lainnya untuk materi yang relevan. Berdasarkan kriteria tertentu, kami memilih 53 publikasi dari total 196 penelitian. Tinjauan ini menganalisis teknik deep learning, modifikasi dataset menggunakan metode prapemrosesan dan segmentasi, serta metrik evaluasi yang digunakan. Selain itu, evaluasi meta-analitik dilakukan pada metrik kinerja untuk mengidentifikasi tren di antara dataset yang berbeda dalam pembangkitan model deep learning. Kami juga secara singkat menguraikan tren baru seperti ensembel dan fungsi blok, terutama dalam CNN, dengan beberapa catatan penting dari perspektif klinis. Masih ada masalah kunci yang perlu diatasi, seperti masalah bias dataset, skalabilitas dalam lingkungan dengan sumber daya terbatas, dan jaminan generalisasi. Metodologi terstruktur diterapkan dalam tinjauan ini dalam bentuk tabel, untuk mengidentifikasi dan mengekstrak tren teknologi deep learning. Ekspedisi sistematis ini diharapkan dapat memicu rasa ingin tahu terhadap kompleksitas operasi deep learning dan memberikan pandangan luas tentang perkembangan teknologi di bidang kecerdasan buatan. Hasilnya akan mendukung langkah-langkah penting di bidang di mana kesehatan dan teknologi saling beririsan.

No Type BC Dataset Model Overview of CNN Blocks Result Author Keterangan
11 Mitotic cell detection ICPR 2012 Mitosis Dataset, ICPR 2014 Mitosis Dataset, AMIDA13 Dataset Fully Convolutional Networks (FCNs) Not reported On the 2014 ICPR mitosis dataset, the model achieved an F-score of 0.575, while on the AMIDA13 dataset, it obtained an F-score of 0.698. PartMitosis: A Partially Supervised Deep Learning Framework for Mitosis Detection in Breast Cancer Histopathology Images
12 Mass and Microcalcifications Mammographic Image Analysis Society (MIAS) A hybrid classifier combining Extreme Learning Machine (ELM) with Glowworm Swarm Optimization (GSO) and Fruit Fly Optimization Algorithm (FOA), hereafter denoted GSO-ELM-FOA, was developed. Reported Acc=0.9915, Prec=1, Sens=0.9791 Detection and classification of breast cancer from digital mammograms using hybrid extreme learning machine classifier
13 Invasive Ductal Carcinoma (IDC) Wisconsin breast cancer (WBC) Optimized Stacking Ensemble Learning (OSEL) model Not reported Acc=0.9945, Pre=0.9900, Rec=0.9800, F-measure=0.9900, Optimized Stacking Ensemble Learning Model for Breast Cancer Detection and Classification Using Machine Learning
14 Malign Breast Cancer Microwave Scattering Dataset from Breast Phantoms for Tumor Detection comprises calibrated experimental measurement data on gel-like phantoms mimicking human tissue, tested at the ERMIAS Laboratory, University of Calabria. A convolutional neural network (CNN) was utilized to estimate dielectric properties (relative permittivity and conductivity) from electromagnetic measurements of breast phantoms, while the quadratic programming-based BIM framework was applied to solve the Reported The combined CNN and quadratic BIM method achieved over 90% accuracy in reconstructing relative permittivity and conductivity within ~15 minutes. Compared to the standard quadratic BIM with error rates of 44?89%, the hybrid approach reduced errors to below 10%. Machine Learning Approach to Quadratic Programming-Based Microwave Imaging for Breast Cancer Detection
15 Malign Breast Cancer The Cancer Genome Atlas (TCGA) dan Genomic Data Commons (GDC) Data Portal Machine learning models such as RF, SVM, NB, KNN, DT, LR, ANN, and deep learning (DL) represent the primary methods frequently employed for breast cancer classification and detection using miRNA expression data in the reviewed studies. Not reported Based on the reviewed studies, Deep Learning (DL) and Artificial Neural Networks (ANN) achieved the highest performance in breast cancer classification and detection using miRNA biomarkers, both reporting accuracies up to 100%, while other methods such as DT (99.12%), KNN (99.2%), RF (AUC 99.5?99.9%), and SVM (AUC 93.8?99.6%) showed strong but slightly lower results. Addressing the Clinical Feasibility of Adopting Circulating miRNA for Breast Cancer Detection, Monitoring and Management with Artificial Intelligence and Machine Learning Platforms
16 Malign Breast Cancer Numerical Breast Phantom Repository milik University of Wisconsin Cross-Disciplinary Electromagnetics Laboratory (UWCEM) The Born Iterative Method (BIM) combined with quadratic programming and a U-Net-based convolutional neural network (CNN). Not reported Acc=0.9540 with an average relative error of 2.4%, the quadratic BIM reconstruction required less than 5 minutes per image, while the trained CNN reduced the reconstruction time to under 5 seconds per image Enhanced Machine Learning Approach for Accurate and Fast Resolution of Inverse Scattering Problem in Breast Cancer Detection
17 Breast Pathology Wisconsin breast cancer (WBC) This study conducted a comparative analysis using Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR), AdaBoost, Decision Tree, and Random Forest. Not reported The study demonstrated that Support Vector Machine (SVM) achieved the best performance, reaching up to 0.977 accuracy with AUC 0.99, outperforming other machine learning models in breast cancer classification. Bio-Imaging-Based Machine Learning Algorithm for Breast Cancer Detection
18 Breast Pathology Wisconsin breast cancer (WBC) Deep features extracted via CNN were classified using a voting ensemble combining Logistic Regression and Stochastic Gradient Descent. Not reported LR+SGD: Acc=1, Prec=1, Rec=1, F1-score=1 Breast Cancer Detection Using Convoluted Features and Ensemble Machine Learning Algorithm
19 Invasive Ductal Carcinoma (IDC) CAMELYON16, CAMELYON17, LocalSentinel (161 slides), LocalAxillary (57 slides), and LocalNegativeAxillary (259 slides). This study utilized a modified DenseNet architecture, retrained with local datasets (LocalSentinel and LocalAxillary) in addition to CAMELYON data, to enhance generalization for whole-slide image analysis. Not reported The deep learning model achieved high performance on CAMELYON16: AUC=0.969, FROC=0.838, but declined on LocalSentinel: AUC=0.929, FROC=0.744) and LocalAxillary (FROC=0.503), while retraining with local data improved performance by ~4% on LocalSentinel and up to 11% (AUC) and 49% (FROC) on LocalAxillary. Generalization of Deep Learning in Digital Pathology Experience in Breast Cancer Metastasis Detection
20 Breast Pathology CBIS-DDSM (Curated Breast Imaging Subset of Digital Database for Screening Mammography) and The Cancer Imaging Archive (TCIA), Kaggle The ensemble machine learning models included Random Forest (RF), eXtreme Gradient Boosting (XGB), Decision Tree (DT), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN). Not reported From these model combinations, several ensemble models were developed, with the best being the RF-XGB-10, which utilized 10 selected features based on the Random Forest feature importance algorithm. This model achieved a testing Acc=0.9805, MCC=0.9727% F1-Score=0.9805, AUC=0.9891 An Effective Ensemble Machine Learning Approach to Classify Breast Cancer Based on Feature Selection and Lesion Segmentation Using Preprocessed Mammograms
Kesimpulan

Ulasan literatur sistematis ini mengkaji penerapan metode machine learning dan deep learning dalam deteksi serta segmentasi kanker payudara dengan menjawab empat pertanyaan penelitian (RQ1-RQ4). Hasil kajian menunjukkan bahwa metode machine learning tradisional masih relevan karena efisiensi komputasi dan kinerjanya yang tinggi pada dataset medis berukuran kecil hingga menengah. Namun, deep learning-khususnya Convolutional Neural Network (CNN) dengan arsitektur seperti U-Net, ResNet, DenseNet, dan EfficientNet-menjadi pendekatan yang paling dominan karena kemampuannya dalam mengekstraksi fitur secara otomatis dan menghasilkan performa yang unggul dalam tugas deteksi dan segmentasi. Pendekatan hybrid dan ensemble terbukti memberikan akurasi yang lebih stabil dan konsisten. Selain itu, kualitas dataset, teknik prapemrosesan, strategi segmentasi, serta modifikasi arsitektur model (termasuk mekanisme perhatian dan pretrained weights) berperan penting dalam meningkatkan akurasi, efisiensi komputasi, dan kemampuan generalisasi model. Meskipun demikian, tantangan seperti bias data, keterbatasan sumber daya komputasi, generalisasi model, serta isu etika dan privasi masih menjadi perhatian utama dalam penerapan klinis.
Sebagai bentuk implementasi dan dukungan terhadap penelitian, dikembangkan sebuah aplikasi web yang berfokus pada edukasi kanker payudara sekaligus mendukung proses penyusunan Systematic Literature Review (SLR). Aplikasi ini memanfaatkan data artikel ilmiah yang diperoleh dari OpenAlex untuk membantu pengguna dalam proses pencarian, pengelolaan, dan analisis literatur secara sistematis. Selain menyediakan informasi edukatif terkait kanker, web ini juga mampu menghasilkan laporan SLR secara otomatis serta mendukung pengunduhan diagram dan laporan PRISMA, sehingga mempermudah peneliti dalam mendokumentasikan alur seleksi literatur secara transparan dan terstandar. Pendekatan ini diharapkan dapat meningkatkan efisiensi penelitian, kualitas pelaporan ilmiah, serta akses terhadap pengetahuan yang berbasis bukti

📈 Tren Penelitian Berdasarkan Jenis (Type BC)