Page 118 - AIH-2-2

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

including patient demographics, radiologist expertise, and (v) Evolving field: The rapid pace of technological
technical variations. LB represents a promising avenue for advancement in both AI and biomarker discovery
breast cancer detection through a minimally invasive blood means that some of the reviewed techniques
test. However, its application in clinical settings is hindered may already be outdated or superseded by newer
by limitations in sensitivity and specificity, particularly methods not captured in this review.
in early-stage diseases where target biomarkers are less (vi) Limited economic analysis: This review does not
abundant. Standardization of techniques and protocols thoroughly assess the cost-effectiveness of the novel
is crucial to ensure consistent and reliable results across detection methods, which is a crucial factor for real-
laboratories. Ongoing research efforts are directed toward world implementation.
enhancing sensitivity through advanced technologies (vii) Potential for overestimation: As noted by the reviewer,
such as digital PCR and next-generation sequencing. the field is currently in an “enthusiastic” phase, which
Furthermore, the integration of LB with existing diagnostic may lead to an overemphasis on the advantages
modalities, such as imaging and AI, holds promise for of these novel methods while underestimating
refining early detection strategies. Furthermore, long non- potential limitations that may become apparent with
coding RNAs and other novel biomarkers are emerging as more extensive clinical application.
potential tools for non-invasive breast cancer detection. (viii) Complexity of breast cancer biology: The review
Nonetheless, challenges regarding result reproducibility may not fully capture the implications of breast
and variable sensitivity across different techniques need to cancer’s heterogeneity, multifocality, and complex
be considered to fully realize their potential. morphology on the effectiveness of these detection

Among the biomarkers studied, miRNA molecules methods across all cancer subtypes and stages.
were the most frequently studied biomarkers in the early
diagnosis of breast cancer, as shown in Table 6. Specifically, 6. Conclusion
miRNA-382 and miRNA-21 have the highest diagnostic Our systematic review of novel methods for early breast
potential to discern between healthy individuals and cancer detection reveals promising advancements in both
breast cancer patients (Table 5). There are also panels CAD systems and LB techniques. While both approaches
of biomarkers that address the limitation of diagnostic show potential, they each have distinct advantages and
sensitivity. 30,34 limitations that warrant careful consideration in their
implementation and future development.
5. Limitations of the study
In examining CAD systems, several significant
It is crucial to acknowledge the limitations inherent in this advantages have emerged. Many of these systems have
systematic review: demonstrated impressive accuracy rates above 90% in
(i) Selection bias: The review is based on published distinguishing between benign and malignant cases. Their
studies available in selected databases (PubMed, ability to integrate with existing imaging technologies has
ACM Digital Library, and SciSpace). This approach proven valuable in enhancing the effectiveness of traditional
may have excluded relevant studies not indexed in mammography and ultrasound screenings. Furthermore,
these databases or unpublished research, potentially AI-assisted diagnosis shows promise in reducing human
introducing selection bias. error by helping to mitigate the variability often seen in
(ii) Publication bias: There may be a tendency for radiologist interpretations. However, CAD systems also
positive results to be published more frequently face notable limitations. Their performance can vary
than negative or inconclusive findings, which could significantly based on the training dataset, potentially
skew the overall assessment of the effectiveness of limiting generalizability across different populations. This
the methods reviewed.
(iii) Heterogeneity of studies: The reviewed studies challenge is compounded by the need for large, diverse
datasets to ensure robust performance across various
employ diverse methodologies, sample sizes, and
outcome measures, making direct comparisons demographic groups. In addition, the integration of CAD
systems into existing clinical workflows presents logistical
challenging and potentially limiting the and acceptance hurdles that must be addressed.
generalizability of conclusions.
(iv) Lack of long-term follow-up: Many of the studies, LB techniques offer their own set of advantages and
particularly those involving novel biomarkers or AI challenges. The non-invasive nature of LB, requiring only
techniques, lack long-term follow-up data, which is a blood draw, significantly reduces patient discomfort
crucial for assessing the true clinical impact of these compared to traditional diagnostic methods. These
methods. techniques show particular promise in early detection,

Volume 2 Issue 2 (2025) 112 doi: 10.36922/aih.4197