Page 119 - AIH-2-2

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

with some biomarkers demonstrating the ability to not only provide accurate predictions but also offer
detect cancer before clinical symptoms appear. The interpretable explanations for their decisions, potentially
comprehensive molecular profiling provided by LB can utilizing techniques such as SHAP values or LIME. This
offer valuable insights into tumor genetics and potential transparency will be essential for building trust among
treatment responses. However, the approach faces several healthcare professionals. In addition, the field requires
limitations, including considerable variability in sensitivity standardization and benchmarking efforts to establish
and specificity across different biomarkers and studies. consistent protocols for LB sample collection, processing,
Standardization challenges persist, with inconsistent and AI-based analysis, along with creating benchmark
protocols potentially affecting result reproducibility across datasets and evaluation metrics for fair comparison of
laboratories. In addition, the limited sensitivity in early- different AI approaches.
stage disease due to the low abundance of biomarkers can
impact detection rates. Clinical implementation represents another key area for
development. AI-powered clinical decision support systems
When comparing these approaches, CAD systems should be developed and evaluated to integrate LB results with
currently appear more immediately applicable to clinical other clinical data, including imaging and patient history,
practice, given their integration with existing imaging to provide comprehensive risk assessments and treatment
modalities and high reported accuracy rates. However, recommendations. Alongside this technical development,
LB, particularly using miRNA panels, shows significant thorough cost-effectiveness analyses are essential. These
potential for future development, especially in detecting studies should evaluate AI-enhanced LB approaches
early-stage cancers that may be missed by imaging alone. against traditional screening methods, considering factors
The most promising approaches likely involve combining such as equipment costs, personnel training, and potential
these methods. For instance, using LB as an initial reductions in unnecessary procedures.
screening tool, followed by AI-enhanced imaging for
confirmation and localization, could leverage the strengths The field would also benefit from focused research
of both techniques while mitigating their limitations. on treatment monitoring and privacy-preserving
computation. AI models should be investigated for
Based on our review, we have identified several critical
areas for future research in AI-enhanced LB for breast their ability to analyze serial LB samples during cancer
cancer detection. First and foremost, there is a pressing treatment, providing early indicators of treatment efficacy
need for large-scale, multi-center studies with diverse or resistance. Furthermore, federated learning approaches
patient populations. These prospective studies should should be explored to enable AI model training across
aim to validate AI models’ performance in analyzing multiple institutions without compromising patient data
LB data, with sample sizes of at least 1000 patients, privacy, facilitating larger-scale studies while addressing
encompassing various breast cancer subtypes and data security concerns.
stages. Alongside this, the integration of multi-omics By focusing research efforts on these specific areas,
data presents a crucial opportunity for advancement. AI we can advance the field of AI-enhanced LB for breast
models should be developed to analyze multiple biomarker cancer detection, potentially leading to more accurate,
types simultaneously, including ctDNA, miRNAs, and cost-effective, and widely applicable screening methods.
proteins, potentially utilizing advanced ML techniques The integration of these various research directions
such as multi-modal DL or ensemble methods to improve could significantly improve our ability to detect and
detection accuracy. monitor breast cancer, ultimately benefiting patient
Longitudinal research represents another vital avenue outcomes. In conclusion, while the varied methodologies
for investigation. Studies should be designed to track and performance metrics across studies make direct
patients over extended periods, ideally 5 years or more, with comparisons challenging, both CAD and LB demonstrate
regular LB sampling. These temporal data could be analyzed significant potential in advancing early breast cancer
using AI to identify early markers of cancer development detection. The future of breast cancer diagnostics likely
or recurrence. In parallel, AI-driven biomarker discovery lies in the strategic integration of these novel approaches
should be pursued using advanced techniques such as with existing screening methods, paving the way for more
unsupervised learning or reinforcement learning, which accurate, accessible, and personalized early detection
may reveal novel biomarkers that traditional analysis strategies.
methods might miss.
Acknowledgments
The development of explainable AI models is
crucial for clinical adoption. These frameworks should None.

Volume 2 Issue 2 (2025) 113 doi: 10.36922/aih.4197