Feature-Centric AI for Breast Cancer Metastasis Analytics: A Systematic Review and Evidence-Based Framework for Feature Engineering and Multimodal Integration

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript examines a broad range of topics across multiple evidence sources, covering radiomics, clinical variables, pathological features, molecular markers, blood-based biomarkers, and multimodal feature integration, with literature drawn from MEDLINE, Scopus, Web of Science, Embase, and IEEE Xplore.
The systematic review methodology of the manuscript followed a structured PRISMA-based approach for study identification, screening, eligibility assessment, quality appraisal, data extraction, and thematic synthesis, with the methodological quality of included studies assessed using MMAT. The manuscript is detailed synthesis, classification, and interpretation of feature-processing and engineering techniques as well as feature types used in AI-based breast cancer metastasis analytics. The review provides detailed observations, including the central role of imaging radiomics, the importance of preprocessing and reproducibility assessment, the frequent use of LASSO and related feature-selection methods, the growing relevance of multimodal integration, and the increasing incorporation of explainable AI. The 5Fs Framework organizes the feature engineering process into five structured stages, from feature extraction to interpretability, and highlights the importance of feature quality, integration, and explainability in AI-based breast metastasis prediction. 

A major reference integrity issue is present. Ref. [86] appears to refer to a study on pancreatic cancer and pancreatitis rather than breast cancer metastasis analytics. The authors should check this Ref. [86] again.

The epidemiological statement reporting approximately 2.3 million new breast cancer diagnoses in 2024 should be corrected to 2022, unless a valid source for the 2024 estimate is provided. Other citation-related statements should also be checked for consistency.

The MMAT quality appraisal should be clarified by explaining how “Yes,” “No,” and “Can’t tell” ratings were translated into "High" or "Moderate Quality" categories.

The 5Fs Framework should be described as an evidence-informed conceptual framework derived from narrative synthesis, rather than as a validated clinical framework.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This systematic review synthesises evidence from 50 empirical studies on feature engineering, multimodal integration, and explainability in AI-driven breast cancer metastasis analytics. The manuscript is well organised and provides a broad overview of current approaches across radiomics, clinical, molecular, biomarker, and pathology-derived data. The proposed 5Fs Framework summarises the reviewed workflows into five feature-engineering stages.

One aspect that I personally find less compelling, although I do not consider it a methodological weakness, is the positioning of the proposed 5Fs Framework. The workflow of feature extraction, preprocessing/transformation, feature selection, integration, and interpretability is already broadly familiar in ML and radiomics research, so its conceptual novelty may be viewed differently by different readers.

Comments regarding general concepts / Specific comment

Lines 206–227, Section 2.5 Quality Appraisal: Please clarify which criteria were assessed directly using the MMAT and which criteria were added specifically for appraisal of AI, radiomics, and machine-learning studies. It would also be helpful to briefly explain how the narrative classifications of “high quality” and “moderate quality” were derived from the MMAT responses together with the supplementary AI-specific considerations.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf