Over the past few years, artificial intelligence (AI) has begun to fulfill its promise of revolutionizing healthcare. In this issue of Haematologica, Clichet and colleagues describe how applying AI to flow cytometry parameters improves diagnostic accuracy in patients suspected for myelodysplastic neoplasms (MDS).1
Since MDS can be challenging to distinguish from benign cytopenias based on standard diagnostic parameters, multiparametric flow cytometry (MFC) has emerged as an additional diagnostic tool.2 However, extensive MFC panels with complex analysis strategies are required to achieve adequate diagnostic accuracy. Employing AI may address this issue, given its potential to detect patterns in complex data.3 While the use of AI in the diagnosis of medical conditions has been a topic of debate, recent advancements demonstrate its benefits in different illnesses such as diabetic retinopathy and breast cancer.3,4 For both diseases, diagnostic AI tools are currently commercially available and approved for clinical use by regulatory bodies.3,4
In the current research, Clichet and colleagues employed an AI model to parameters derived from the MFC data needed for the Ogata score. The Ogata score is the most used MFC score for MDS diagnostics and requires the assessment of only two cell surface proteins: CD34 and CD45.5 Although the Ogata score is useful and easy to implement, it has limited sensitivity, ranging from 34 to 76%. Clichet and colleagues illustrate that the use of an elastic net AI model resulted in a simple but accurate diagnostic model that only requires four parameters. This model obtained a sensitivity of 91.8% and a specificity of 92.5%, and was validated in an external cohort of 89 patients, illustrating its multi-center potential. One of the explanations for the increase of sensitivity compared with the Ogata score is presumably that this model uses continuous parameters (instead of a fixed cut-off like in the Ogata score) and assigns a weight to each parameter based on its relevance for MDS diagnosis.
A major challenge for AI in healthcare is translating a successful AI model into widespread clinical implementation. Before implementing the model developed by Clichet and colleagues, two initial steps must be taken: extensive multi-center validation and harmonization. Even though most data will be readily available at many locations, such as those collaborating within the European Leuke-miaNet working group, harmonization of data acquisition and manual gating strategies of MFC data is crucial. Alternatively, manual gating could be completely replaced by automated analyses, which removes inter-operator variation but requires higher levels of harmonization of data acquisition.6,7
Implementing AI-based diagnostics is also challenging in itself, and lack of technical expertise and funding are commonly reported as major obstacles.8 However, there are examples of innovative solutions that can help facilitate the adoption of AI models in healthcare. One such example is the recently introduced Molecular International Prognostic Scoring System for Myelodysplastic Syndromes (IPSS-M) risk stratification model for MDS, which can be easily employed using a web-based tool that is accessible to all.9 The availability of a similar tool for the model developed by Clichet and colleagues would make it easier to use and facilitate widespread adoption in clinical settings.
Future development and implementation of AI-based tools for the diagnosis of MDS will largely depend on the availability of high-quality data.8 The previously mentioned successful studies on AI in diabetic retinopathy and breast cancer cover thousands of patients, spread over multiple centers and continents.3,4 To fully harness the potential of AI in the diagnosis of MDS, it is crucial that the MDS community joins forces to continue to build comprehensive and diverse databases, such as those managed by the European LeukemiaNet MDS (EUMDS) registry and the MDSRight Consortium. To further improve diagnostic accuracy, these databases should be expanded with data offering high diagnostic potential when applying AI, such as morphological images, mutational data, and data from novel tools such as label-free cyto-metry.10,11
Overall, we commend Clichet and colleagues for developing an elegant diagnostic model for MDS. By establishing high-quality databases and clear guidelines on how to implement AI-based diagnostic tools effectively, we can further advance the diagnosis of MDS through AI.
- Received March 30, 2023
- Accepted April 7, 2023
No conflicts of interest to disclose.
All authors contributed equally to the content of this editorial and approved the final submitted manuscript.
- Clichet V, Lebon D, Chapuis N. Artificial intelligence to empower diagnosis of myelodysplastic syndromes by multiparametric flow cytometry. Haematologica. 2023; 108(9):2435-2443. Google Scholar
- Brunner AM, Leitch HA, van de Loosdrecht AA, Bonadies N. Management of patients with lower-risk myelodysplastic syndromes. Blood Cancer J. 2022; 12(12):166. Google Scholar
- Kann BH, Hosny A, Aerts HJWL. Artificial intelligence for clinical oncology. Cancer Cell. 2021; 39(7):916-927. Google Scholar
- Grauslund J. Diabetic retinopathy screening in the emerging era of artificial intelligence. Diabetologia. 2022; 65(9):1415-1423. Google Scholar
- Ogata K, Della Porta MG, Malcovati L. Diagnostic utility of flow cytometry in low-grade myelodysplastic syndromes: a prospective validation study. Haematologica. 2009; 94(8):1066. Google Scholar
- Duetz C, Bachas C, Westers TM, van de Loosdrecht AA. Computational analysis of flow cytometry data in hematological malignancies: future clinical practice?. Curr Opin Oncol. 2020; 32(2):162-169. Google Scholar
- Duetz C, Van Gassen S, Westers TM. Computational flow cytometry as a diagnostic tool in suspected-myelodysplastic syndromes. Cytometry A. 2021; 99(8):814-824. Google Scholar
- He J, Baxter SL, Xu J, Xu J, Zhou X, Zhang K. The practical implementation of artificial intelligence technologies in medicine. Nat Med. 2019; 25(1):30-36. Google Scholar
- Bernard E, Tuechler H, Greenberg PL. Molecular International Prognostic Scoring System for Myelodysplastic Syndromes. NEJM Evidence. 2022; 1(7):EVIDoa2200008. Google Scholar
- Kimura K, Tabe Y, Ai T. A novel automated image analysis system using deep convolutional neural networks can assist to differentiate MDS and AA. Sci Rep. 2019; 9(1):13385. Google Scholar
- Brück OE, Lallukka-Brück SE, Hohtari HR. Machine learning of bone marrow histopathology identifies genetic and clinical determinants in patients with MDS. Blood Cancer Discov. 2021; 2(3):238-249. Google Scholar
Figures & Tables
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.