Page 117 - AIH-2-2

Artificial Intelligence in Health AI in early breast cancer diagnosis: A review

Computer-aided early diagnosis of breast cancer has technique of mitotic cells in blood as cancer biomarkers.
20
become a pivotal area of research, leveraging advancements The study reported varying precision metrics on different
in ML and DL to enhance diagnostic accuracy and early datasets.
intervention. Traditional detection methods, such as Sharifi et al. analyzed two microarray datasets
34
mammograms, have been significantly improved by ML (GSE106817 and GSE113486) to investigate miRNAs in the
algorithms that can identify subtle changes in breast tissue serum of breast cancer patients and healthy individuals.
34
indicative of cancer, even before tumors are visible on The study identified the combined occurrence of miR-
mammograms. 92a-3p, miR-23b-3p, and miR-191-5p was the most
Various ML techniques, including SVM, kNN, and promising pattern to distinguish breast cancer patients
Naïve Bayes, have shown high accuracy in classifying breast from healthy controls with 0.89 sensitivity, 0.96 specificity,
cancer, with SVM achieving up to 98.24% accuracy. DL and an area under the curve (AUC) of 0.98. These
21
models, especially CNNs, have further advanced this field miRNAs represent new biomarkers for early detection of
by automatically extracting features from mammogram breast cancer. Kim et al. studied miRNAs from tumor-
35
images, achieving training accuracies up to 99% and derived extracellular vesicles (EVs). The study reported
35
validation accuracies up to 96%. The integration of CAD the fabrication of a novel microfluidic device that can
22
systems utilizing these ML and DL models has proven promptly and selectively separate cancerous EVs. Among
effective in assisting radiologists by providing a second the candidate miRNAs studied from these EVs, four
opinion, thus reducing false positives and negatives. 21 miRNAs (miR‐9, miR‐16, miR‐21, and miR‐429) were
elevated in early‐stage breast cancer patients as opposed
Despite these advancements, challenges such as to healthy individuals, exhibiting high sensitivities of 0.90,
overfitting in CNNs and the need for standardized data 0.86, 0.88, and 0.84 in each subtype of early‐stage breast
collection and analysis processes persist. Overall, the cancer. This proposed methodology not only diagnosed
integration of AI in breast cancer detection not only early-stage breast cancer but also determined the cancer
enhances early diagnosis but also supports clinical subtype.
decision-making, potentially saving countless lives by
37
enabling timely and accurate treatment. Lin et al. studied levels of circular RNA (circRNA) in
259 samples (144 breast cancer patients and 115 controls)
3.2.2. Early breast cancer diagnosis by biomarkers and developed classifiers from 20 candidate circRNA with
obtained from LB elevated levels, achieving an AUC of 0.83. The significance
Fourteen studies focused on exploring biomarkers as of this study lies in the potential use of circRNAs in plasma
potential tools for non-invasive and early diagnosis of EVs in the diagnosis and management of breast cancer.
breast cancer. Table 6 provides an overview of the types of 4. Discussion
biomarkers isolated in these articles, and Table 5 further
expands on the diagnostic potential of each type for clinical Various architectures (e.g., U-Net, ResNet) have been
applications. successfully applied in breast cancer detection, achieving
high accuracy and precision in classification tasks.
As discussed in the previous CAD results section,
Mahmood et al. proposed a counting and detection Notable methods include combining residual and U-Net
20
architectures, ensemble models, and transfer learning.
Traditional ML models such as kNN, LDA, and PNN
Table 6. Number of eligible articles identified based on types have demonstrated high success rates. New approaches,
of biomarkers (n=14) such as feature ranking and particle swarm optimization,
Biomarker type Number of Percentage of have further enhanced detection accuracy. It is common
articles articles for these methods to achieve high accuracy (above 90%)
microRNA 8 57.14 in distinguishing between benign and malignant cases.
Plasma metabolites 1 7.1 The use of advanced models and techniques significantly
Circular RNA 1 7.1 reduces false positives and improves sensitivity and
specificity. These results indicate that leveraging advanced
tRNA-derived fragments 1 7.1
computational techniques in CAD systems significantly
Receptor-activator proteins 1 7.1 enhances the early diagnosis of breast cancer.
Amino acid 1 7.1 While current breast cancer screening methods, such
Protein 1 7.1 as mammography and ultrasound, offer valuable tools for
Mitotic cells 1 7.1 detection, their effectiveness can be limited by factors,

Volume 2 Issue 2 (2025) 111 doi: 10.36922/aih.4197