Molecular Pathology, Artificial Intelligence, and New Technologies in Hematologic Diagnostics: Translational Opportunities and Practical Considerations

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This review is very interesting, as it states the following: AI models in molecular hematopathology are assisting in mutation interpretation, genetic risk stratification, and correlation of morphological and genomic findings. AI is beginning to change the way hematopathology and molecular diagnostics are performed. Successful translational research depends on disease-specific validation, development of multimodal models that comply with ICC and WHO frameworks, and laboratory governance that maintains expert oversight. However, the following points would be even better if they were added: (1) What exactly is the high concordance rate between digital morphology analyzers and manual microscopy? Does it vary depending on deep learning? (2) Was the manual microscopy performed by a specialized technician or a specialist physician? (3) How many samples were tested for comparison? (4) Benefits of introducing AI: 1. It helps pathologists and technicians deal with strict time constraints. 2. AI-based systems can assist with data interpretation by prioritizing variants, identifying complex or unexpected genetic lesion combinations, and ensuring alignment with WHO and ICC classification frameworks.

Author Response

Response to Reviewer 1:

We thank the reviewers for the assessment of our manuscript and for the constructive suggestions. We have revised the manuscript accordingly and addressed each point below.

Comment 1: What exactly is the high concordance rate between digital morphology analyzers and manual microscopy? Does it vary depending on deep learning?

Response: We agree that quantitative clarification strengthens the manuscript. We have added specific concordance ranges reported in the literature and clarified that performance varies depending on platform, dataset composition, and whether AI-assisted classification is used.

Manuscript change: A paragraph has been added in Section 3.2 Digital Morphology and AI in Peripheral Blood and Bone Marrow describing concordance rates (typically 85–98% depending on cell category and study design) and highlighting that AI-enabled preclassification generally improves concordance and efficiency.

(2) Was the manual microscopy performed by a specialized technician or a specialist physician?

Response: We thank the reviewer for this important clarification. In most validation studies, manual differentials are performed by experienced laboratory technologists and/or hematopathologists, often with consensus adjudication. This clarification has been added.

Manuscript change: We now explicitly state that manual microscopy comparisons were performed by experienced technologists and hematopathologists, as specified by the study design.

(3) How many samples were tested for comparison?

Response: We have added representative study sample sizes from cited evaluations (ranging from ~100 to >1,000 smears depending on platform and validation cohort).

Manuscript change: Sample size ranges have been incorporated into the digital morphology section.

(4) Benefits of introducing AI: 1. It helps pathologists and technicians deal with strict time constraints. 2. AI-based systems can assist with data interpretation by prioritizing variants, identifying complex or unexpected genetic lesion combinations, and ensuring alignment with WHO and ICC classification frameworks.

Response: We have expanded the discussion of practical clinical benefits, including workflow efficiency, diagnostic triage, and variant prioritization aligned with WHO/ICC frameworks.

Manuscript change: A dedicated paragraph describing workflow efficiency, time reduction, and molecular variant prioritization has been added in Sections 3.2 and 3.4.

Reviewer 2 Report

Comments and Suggestions for Authors The manuscript “Molecular Pathology, Artificial Intelligence, and New Technologies in Hematologic Diagnostics: Translational Opportunities and Practical Considerations” addresses a very timely topic that will undoubtedly have an important impact on the future of hematology. It brings together recent advances in molecular techniques, digital tools, and AI applications in diagnostic hematology. The topic is interesting and relevant, but the manuscript would clearly benefit from a narrower focus and deeper technical detail. Some suggestions:

• At present, the review covers a very broad spectrum and often stays on a general level, with limited explanation of how the different technologies actually work or are applied in practice; i.e., I would like to see more data on whether learning is supervised (with labeled data) or unsupervised (automatic patterns).
• The first sections concentrate on AML, while later parts jump to other hematologic diseases without a clear rationale or transition, which gives an impression of fragmentation. It might be more effective to organize the text around specific technologies or clinical applications, which would make the discussion more coherent and easier to follow.
• Although the title refers to the whole of hematology, I miss a bit more variability and breadth, talking about other pathologies, such as multiple myeloma.
• The references are mostly recent and appropriate, but some of the main assertions are not fully supported (for example, lines 242-244 about molecular MRD). Strengthening the bibliography where conclusions are drawn would make the argumentation more solid.
• There is also a numbering error in the citation list that should be corrected (references number 23-26 are after 34-37 and that is a mistake).
• The inclusion of figures would substantially improve the paper. Visual material comparing technologies or summarizing workflows could make the content clearer and more instructive for readers. As it stands, the text alone feels too abstract for such a technical subject.

In summary, the review deals with a highly relevant and forward-looking area, but it needs a major revision focused on better structure, more in-depth analysis, and stronger visual and bibliographic support.

Author Response

Reviewer 2:

• At present, the review covers a very broad spectrum and often stays on a general level, with limited explanation of how the different technologies actually work or are applied in practice; i.e., I would like to see more data on whether learning is supervised (with labeled data) or unsupervised (automatic patterns).

Response: We agree and have added additional methodological detail describing supervised, unsupervised, and deep learning approaches across morphology, flow cytometry, and molecular applications.

Manuscript change: A new subsection “3.0” explaining learning paradigms and their clinical implications has been incorporated.

• The first sections concentrate on AML, while later parts jump to other hematologic diseases without a clear rationale or transition, which gives an impression of fragmentation. It might be more effective to organize the text around specific technologies or clinical applications, which would make the discussion more coherent and easier to follow.

Response: We appreciate this observation and have reorganized the manuscript to emphasize technology centered structure rather than disease centered flow, improving continuity.

Manuscript change: Transitions were added and sections were reorganized to highlight modality based discussion.

• Although the title refers to the whole of hematology, I miss a bit more variability and breadth, talking about other pathologies, such as multiple myeloma.

Response: We have expanded the discussion to include additional hematologic diseases.

Manuscript change: A dedicated paragraph has been added in section 3.4.

• The references are mostly recent and appropriate, but some of the main assertions are not fully supported (for example, lines 242-244 about molecular MRD). Strengthening the bibliography where conclusions are drawn would make the argumentation more solid.

Response: We strengthened the molecular MRD discussion by adding references to recent consensus guidelines and reviews and clarifying that MRD interpretation must be performed longitudinally and in a clinical context.

• There is also a numbering error in the citation list that should be corrected (references number 23-26 are after 34-37 and that is a mistake).

Response: We thank the reviewer for identifying this error. The reference numbering has been corrected and verified.

• The inclusion of figures would substantially improve the paper. Visual material comparing technologies or summarizing workflows could make the content clearer and more instructive for readers. As it stands, the text alone feels too abstract for such a technical subject.

Response: We thank the reviewer for this very helpful comment. We have added Figure 1 presenting a conceptual overview of the integrated AI enabled hematologic diagnostic workflow, which provides a more structured visual synthesis. Thank you again.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I think the authors have improved the manuscript and taken my suggestions into account.