1. Title of the Publication Evolutionary design of explainable algorithms for biomedical image segmentation ----------------------------------------------------------------------------------------------------------------------- 2. Author Information Kévin Cortacero1,2,3 - kevin.cortacero@inserm.fr Brienne McKenzie1,2,3 - brienne.mckenzie@inserm.fr Sabina Müller1,2,3 - sabina.muller@inserm.fr Roxana Khazen1,2,3 - roxana.khazen@inserm.fr Fanny Lafouresse1,2,3 - fanny.lafouresse@inserm.fr Gaëlle Corsaut1,2,3 - gaelle.corsaut@inserm.fr Nathalie Van Acker4 - VanAcker.Nathalie@iuct-oncopole.fr François-Xavier Frenois4 - fx.frenois@imagin-labs.net Laurence Lamant4 - Lamant.Laurence@iuct-oncopole.fr Nicolas Meyer5 - nicolas.meyer@iuct-oncopole.fr Béatrice Vergier6,7 - beatrice.vergier@chu-bordeaux.fr Dennis G. Wilson8 - Dennis.WILSON@isae-supaero.fr Hervé Luga8 - herve.luga@irit.fr Oskar Staufer9 - oskar.staufer@mpimf-heidelberg.mpg.de Michael L. Dustin9 - michael.dustin@kennedy.ox.ac.uk Salvatore Valitutti1,2,3,4 - salvatore.valitutti@inserm.fr Sylvain Cussat-Blanc - cussat@irit.fr 1. Institut National de la Santé et de la Recherche Médicale (INSERM) UMR1037, Centre de Recherche en Cancérologie de Toulouse (CRCT), Toulouse, France 2. Centre National de la Recherche Scientifique (CNRS) UMR5071, Toulouse, France 3. University of Toulouse III - Paul Sabatier, Toulouse, France 4. Department of Pathology, Institut Universitaire du Cancer-Oncopole de Toulouse (IUCT), 5. Department of Dermatology, IUCT, Toulouse, France 6. Service de Pathologie, Centre Hospitalier Universitaire de Bordeaux, Bordeaux, France 7. INSERM UMR1053 -UMR BaRITOn, Université de Bordeaux, Bordeaux, France 8. University of Toulouse - Institut de Recherche en Informatique de Toulouse (IRIT) - UMR5505, Artificial and Natural Intelligence Toulouse Institute, Toulouse, France 9. Kennedy Institute of Rheumatology (KIR), Nuffield Department of Orthopedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, UK ----------------------------------------------------------------------------------------------------------------------- 3. Corresponding Author Sylvain Cussat-Blanc cussat@irit.fr ----------------------------------------------------------------------------------------------------------------------- 4. Paper Abstract An unresolved issue in contemporary biomedicine is the overwhelming number and diversity of complex images that require annotation, analysis and interpretation. Recent advances in Deep Learning have revolutionized the field of computer vision, creating algorithms that compete with human experts in image segmentation tasks. However, these frameworks require large human-annotated datasets for training and the resulting “black box” models are difficult to interpret. In this study, we introduce Kartezio, a modular Cartesian Genetic Programming-based computational strategy that generates fully transparent and easily interpretable image processing pipelines by iteratively assembling and parameterizing computer vision functions. The pipelines thus generated exhibit comparable precision to state-of-the-art Deep Learning approaches on instance segmentation tasks, while requiring drastically smaller training datasets. This Few-Shot Learning method confers tremendous flexibility, speed, and functionality to this approach. We then deploy Kartezio to solve a series of semantic and instance segmentation problems, and demonstrate its utility across diverse images ranging from multiplexed tissue histopathology images to high resolution microscopy images. While the flexibility, robustness and practical utility of Kartezio make this fully explicable evolutionary designer a potential game-changer in the field of biomedical image processing, Kartezio remains complementary and potentially auxiliary to mainstream Deep Learning approaches. ----------------------------------------------------------------------------------------------------------------------- 5. Criteria that the author claims that the work satisfies (A) The result was patented as an invention in the past, is an improvement over a patented invention, or would qualify today as a patentable new invention. (B) The result is equal to or better than a result that was accepted as a new scientific result at the time when it was published in a peer-reviewed scientific journal. (D) The result is publishable in its own right as a new scientific result independent of the fact that the result was mechanically created. (E) The result is equal to or better than the most recent human-created solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions. ----------------------------------------------------------------------------------------------------------------------- 6. Statement Why the Results Satisfy the Criteria In this work, we used genetic programming to discover pipelines of computer vision functions for cell segmentation that are competitive with human-designed and deep learning methods. Many methods have been proposed for cell segmentation, from manual labeling [1] to hand-crafted pipelines [2, 3], and, more recently, deep learning methods [4]. This is a long-standing problem which has a succession of increasingly better human-created solutions [1, 2, 3], and the result of our work outperforms the most recent human-created solution (E). The evolved pipelines are competitive with the deep learning method Cellpose [4] and outperform the deep learning method Mask R-CNN [5] on this task, both of which were accepted as new scientific results when published (B). There were many pipelines evolved in this work, and they could be both published in their own right as a new scientific result, similar to human-crafted pipelines [2, 3], (D). The evolved pipeline would qualify as a patentable new invention (A), similar to [6], and the discovery pipeline applied to biomedical image analysis has been submitted as a patent application [7]. [1] Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012) [2] Sommer, C., Straehle, C., Koethe, U. & Hamprecht, F. A. Ilastik: interactive learning and segmentation toolkit. IEEE International Symposium on Biomedical Imaging, 230–233 (2011). [3] McQuin, C. et al. Cellprofiler 3.0: next-generation image processing for biology. PLoS Biology 16, e2005970 (2018). [4] Stringer, C., Wang, T., Michaelos, M. & Pachitariu, M. Cellpose: a generalist algorithm for cellular segmentation. Nat. Methods 18, 100–106 (2021). [5] He, Kaiming, et al. "Mask R-CNN." Proceedings of the IEEE international conference on computer vision. 2017. [6] Wilkinson, Jr., D. (2024). Detecting cells of interest in large image datasets using artificial intelligence (U.S. Patent No. 11,989,960). U.S. Patent and Trademark Office. https://ppubs.uspto.gov/dirsearch-public/print/downloadPdf/11989960 [7] Cortacero et al. (2024). EP 22307041.8. Under review. ----------------------------------------------------------------------------------------------------------------------- 7. Full Citation Cortacero, K., McKenzie, B., Müller, S. et al. Evolutionary design of explainable algorithms for biomedical image segmentation. Nat Commun 14, 7112 (2023). https://doi.org/10.1038/s41467-023-42664-x ----------------------------------------------------------------------------------------------------------------------- 8. Prize Money Breakdown We will use the distribution agreed on the patent to split the prize money: Kevin Cortacero: 25% Brienne McKenzie: 7% Hervé Luga: 9% Dennis Wilson: 9% Salvatore Valitutti: 25% Sylvain Cussat-Blanc: 25% ----------------------------------------------------------------------------------------------------------------------- 9. A Statement Indicating Why this Entry Could Be the "Best" Kartezio, our modular Cartesian Genetic Programming-based strategy, generates fully transparent and interpretable image processing pipelines. These pipelines are not only competitive with state-of-the-art Deep Learning methods but also require significantly smaller training datasets. This is particularly beneficial in biomedical applications where obtaining large, annotated datasets can be challenging and expensive. Our results have demonstrated exceptional performance on a variety of complex biomedical images, from multiplexed tissue histopathology to high-resolution microscopy images. The precision of our method matches or surpasses leading Deep Learning approaches like Cellpose, StarDist and Mask R-CNN, showcasing its robustness and effectiveness. The modular and interpretable nature of Kartezio ensures that it can be easily adapted and reused for various biomedical imaging challenges. Our approach has been peer-reviewed and published in Nature Communications, attesting to its scientific validity and potential for widespread adoption in the biomedical community. Kartezio's ability to deliver explainable and high-performance image segmentation with minimal data requirements opens new avenues for research and application in biomedical image analysis. By reducing the dependency on large annotated datasets, our method can accelerate the development and deployment of advanced imaging solutions in resource-constrained settings. Moreover, the transparency and interpretability of the pipelines foster trust and facilitate collaboration between computational scientists and biomedical researchers. The successful application of Kartezio in safety-critical biomedical imaging tasks has the potential to garner significant attention from both the academic community and industry. Our work represents a significant advancement in biomedical image processing, combining the strengths of evolutionary algorithms with the need for transparency and efficiency in medical applications. We believe that the unique advantages and broad applicability of Kartezio make it the best entry for the Humies. ----------------------------------------------------------------------------------------------------------------------- 10. Evolutionary Computation Type Genetic Programming ----------------------------------------------------------------------------------------------------------------------- 11. Publication Date November 6, 2023