1. Title of the publications Article A: Neuro-symbolic interpretable AI for automatic COVID-19 patient-stratification based on standardised lung ultrasound data Article B: Multi-objective automatic analysis of lung ultrasound data from COVID-19 patients by means of deep learning and decision trees 2. Author Information Name: Leonardo Lucio Custode Institution: Dept. of Information Engineering and Computer Science, University of Trento, Trento, TN, Italy Email: leonardo.custode@unitn.it Phone: +39 388 176 7536 Name: Federico Mento Institution: Dept. of Information Engineering and Computer Science, University of Trento, Trento, TN, Italy Email: federico.mento@unitn.it Name: Sajjad Afrakhteh Institution: Dept. of Information Engineering and Computer Science, University of Trento, Trento, TN, Italy Email: sajjad.afrakhteh@unitn.it Phone: +30 347 780 4263 Name: Francesco Tursi Institution: UOS Pneumologia di Codogno, ASST Lodi, Lodi, Italy Email: francesco.tursi@asst-lodi.it Name: Andrea Smargiassi Institution: Dept. of Medical and Surgical Sciences, Fondazione Policlinico Universitario A. Gemelli IRCCS, Rome, Italy Email: andrea.smargiassi@policlinicogemelli.it Name: Riccardo Inchingolo Institution: Dept. of Medical and Surgical Sciences, Fondazione Policlinico Universitario A. Gemelli IRCCS, Rome, Italy Email: riccardo.inchingolo@policlinicogemelli.it Phone: +39 349 842 6623 Name: Tiziano Perrone Institution: Dept. of Internal Medicine, IRCCS San Matteo, Pavia, Italy Email: tiziano.perrone@gavazzeni.it Name: Giovanni Iacca Institution: Dept. of Information Engineering and Computer Science, University of Trento, Trento, TN, Italy Email: giovanni.iacca@unitn.it Phone: +39 0461 285220 Name: Libertario Demi Institution: Dept. of Information Engineering and Computer Science, University of Trento, Trento, TN, Italy Email: libertario.demi@unitn.it Phone: +39 0461 283942 3. Corresponding Author Leonardo Lucio Custode, leonardo.custode@unitn.it 4. Paper Abstracts Paper A: In the current pandemic, being able to efficiently stratify patients depending on their probability to develop a severe form of COVID-19 can improve the outcome of treatments and optimize the use of the available resources. To this end, recent studies proposed to use deep-networks to perform automatic stratification of COVID-19 patients based on lung ultrasound (LUS) data. In this work, we present a novel neuro-symbolic approach able to provide video-level predictions by aggregating results from frame-level analysis made by deep-networks. Specifically, a decision tree was trained, which provides direct access to the decision process and a high-level explainability. This approach was tested on 1808 LUS videos acquired from 100 patients diagnosed as COVID-19 positive by a RT-PCR swab test. Each video was scored by LUS experts according to a 4-level scoring system specifically developed for COVID-19. This information was utilised for both the training and testing of the algorithms. A five-folds cross-validation process was utilised to assess the performance of the presented approach and compare it with results achieved by deep-learning models alone. Results show that this novel approach achieves better performance (82% of mean prognostic agreement) than a threshold-based ensemble of deep-learning models (78% of mean prognostic agreement) Paper B: COVID-19 raised the need for automatic medical diagnosis, to increase the physicians’ efficiency in managing the pandemic. Among all the techniques for evaluating the status of the lungs of a patient with COVID-19, lung ultrasound (LUS) offers several advantages: portability, cost-effectiveness, safety. Several works approached the automatic detection of LUS imaging patterns related COVID- 19 by using deep neural networks (DNNs). However, the decision processes based on DNNs are not fully explainable, which generally results in a lack of trust from physicians. This, in turn, slows down the adoption of such systems. In this work, we use two previously built DNNs as feature extractors at the frame level, and automatically synthesize, by means of an evolutionary algorithm, a decision tree (DT) that aggregates in an interpretable way the predictions made by the DNNs, returning the severity of the patients’ conditions according to a LUS score of prognostic value. Our results show that our approach performs comparably or better than previously reported aggregation techniques based on an empiric combination of frame-level predictions made by DNNs. Furthermore, when we analyze the evolved DTs, we discover properties about the DNNs used as feature extractors. We make our data publicly available for further development and reproducibility 5. Competition Criteria 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. C: The result is equal to or better than a result that was placed into a database or archive of results maintained by an internationally recognized panel of scientific experts. D: The result is publishable in its own right as a new scientific result independent of the fact that the result was mechanically created. F: The result is equal to or better than a result that was considered an achievement in its field at the time it was first discovered. G: The result solves a problem of indisputable difficulty in its field. 6. Why our results satisfy the competition criteria B: The method sets a new state of the art in the field. It improves performance over previous works that used only Deep Neural Networks or a hand-crafted method for combining their predictions [1, 2, 3]. C: The results obtained by our paper outperform previously-published results [1, 2, 3]. D: The result, which is a learned Decision Tree that aggregates the predictions of two existing neural networks is publishable as a novel results that outperform another hand-made system [3], regardless of it being produced by a machine. F: The results obtained by our method outperform the ones obtained in a previous work [3], which was able to substantially increase performance w.r.t. the baseline networks. G: Aggregating Deep Neural Networks is a hard task, because to do that one has to understand what is the inner reasoning performed by the network. Our method solves this problem in a data-driven manner, finding DTs that exploit the biases of the two networks to improve the performance of the overall system. [1] Roy, Subhankar, et al. "Deep learning for classification and localization of COVID-19 markers in point-of-care lung ultrasound." IEEE transactions on medical imaging 39.8 (2020): 2676-2687. [2] Roshankhah, Roshan, et al. "Investigating training-test data splitting strategies for automated segmentation and scoring of COVID-19 lung ultrasound images." The Journal of the Acoustical Society of America 150.6 (2021): 4118-4127. [3] Mento, Federico, et al. "Deep learning applied to lung ultrasound videos for scoring COVID-19 patients: A multicenter study." The Journal of the Acoustical Society of America 149.5 (2021): 3626-3634. 7. Full Citation Custode, Leonardo Lucio et al., "Neuro-symbolic interpretable AI for automatic COVID-19 patient-stratification based on standardised lung ultrasound data", Proceedings of Meetings on Acoustics 46, 2022, Acoustical Society of America Custode, Leonardo Lucio et al., "Multi-objective automatic analysis of lung ultrasound data from COVID-19 patients by means of deep learning and decision trees", Applied Soft Computing, Vol. 133, 2023, Elsevier 8. Prize Breakdown Any prize should be divided equally among the co-authors. 9. Statement stating why this entry should be considered as the "best" The COVID-19 pandemic had a huge impact on our society, infecting about the 10% of the world's population, and leading to about 7 million deaths worldwide [1]. This pandemic induced huge stress on the health systems throughout the world. In fact, often the hospitals' ICUs were saturated, and they were not able to accept new patients, which had disastrous consequences. Especially at the beginning of the pandemic, physicians had to manually assess the health conditions of the lungs of the patients, which occupied a significant portion of their working time. To solve this problem, several approaches have been proposed to automate the assessment of the conditions of the patients' lungs, using deep neural networks. However, most of them could not be applied in hospitals, due to the fact that they could not be validated by physicians. Moreover, the performance of such approaches was not always satisfactory, as these systems were trained on very noisy labels: physicians' subjective evaluation of the patient. In our works, we introduced an explainable component in the system, i.e., decision trees (DTs) trained by means of Grammatical Evolution, equipped with the NSGA-II selection operator. Our approach introduced significant improvements to the existing systems. Firstly, it improved performance both over each single neural network, and over a threshold-based ensemble of the deep networks. Secondly, our approach turned a fully black-box model into a semi-transparent one, which allowed for the partial understanding of the flow of information in the system. This enables validation from the physicians, who can understand the rationale of each decision based on the suggestions of two experts (i.e., the deep networks). Moreover, it enables debugging, allowing for the continuous improvement of the system, which is crucial in the medical domain. Finally, it is important to note that, while the whole system is not interpretable, we can see the resulting DT as an explanation of the flaws of each of the baseline neural networks. In fact, interpreting the flow of information, we were able to understand the most important biases of each of the neural networks. [1] https://covid19.who.int/ 10. Evolutionary Methods used GE (Grammatical Evolution) with NSGA-II. 11. Date of publication Paper A: The paper was published in May 2022. Paper B: The paper was published in January 2023.