By: Luba Gloukhova, Conference Chair, Deep Learning World
In anticipation of his upcoming presentation at Deep Learning World Livestream, May 24-28, 2021, we asked Giovanni Turra, Computer Vision, Machine Learning and Deep Learning Engineer at Copan Group S.p.a., a few questions about their deployment of predictive analytics. Catch a glimpse of his presentation, Deep Learning of Microbiological Analysis inside Full Laboratory Automations, and see what’s in store at the DLW conference.
Q: In your work with deep learning, what do you model (i.e., what is the dependent variable, the behavior or outcome your models predict)?
A: I am working on WASPLab (Laboratory Automations) for microbiological laboratories. Specifically, I develop and apply computer vision and data analysis approaches to elaborate input data (and dataset) like pictures of Petri dishes (or similar) and non-sensitive data, related to analysis from the same sample (or patient). Output results generated can be quite heterogenous and they depend from the type of media (if it allows the growth only of specific pathogens, if it even paints some of them in a specific way, etc…) and sample analyzed.
Generally, our models detect whether something has grown, estimate its number, until identifying which pathogen (or pathogens) is being analyzed and to determine the presence of specific features.
We combine those results with others from different sources, to generate a conclusive result that is useful and usable by microbiologists.
Q: How does deep learning deliver value at your organization – what is one specific way in which model outputs actively drive decisions or operations?
A: Implementation of deep learning techniques began in 2013 and since then they increased their success thanks to the foresight of managers. They have understood their potentials and appreciated the results. Deep learning has allowed at the same time to obtain better performance on fields already proceeding and to open new fields of microbiological analysis and automation that previously had not been considered due to too much complexity and lack of suitable tools.
To give an example, the possibility of identifying specific bacterial species from a set of more than 50 different options required techniques and applications capable of being independent from human evaluation, robust to the variability of an organic structure and capable of generating in a limited time a result that is as accurate as possible without dangerous wrong identifications.
Q: Can you describe a quantitative result, such as the performance of your model or the ROI of the model deployment initiative?
A: An example is by J. Bayette et al. “Evaluation of PhenoMATRIX ™ expert system for the segregation of urine specimens on CHROMIDR CPSE Elite”. In this study, the average decision time per sample analyzed, using one of our software developed with deep learning systems (within our Artificial Intelligence suite called PhenoMATRIX ™) went from 100 s (with a totally manual laboratory workflow) at 1.5 s per sample (with the automated version). PhenoMATRIX ™ reached this amazing result without false negative mistakes and with a specificity score greater than 96% in an extremely complex environment for the microbiological lab.
Q: What surprising discovery or insight have you unearthed in your data?
A: Technology should not be an alternative to microbiologists and technicians but as a complement to them. In this sense, even today, there are conditions in which the human behaves better (when it is necessary to combine local and global knowledge for example) while in other cases the software can combine the presence of features that also they did not seem very useful to us technicians.
We are lucky to be able to apply our approaches on a dataset acquired over the course of several years thanks to the collaboration of many laboratories that already used our products and automations daily: despite the high degree of standardization implemented, growth behavior (of bacteria) PhenoMATRIX™ processes daily remains extremely variable and for this reason our tools must be robust and capable to evaluate situations they have never seen necessarily.
Q: What excites you most about the field of deep learning today?
A: For sure, the variety of areas where can be applied. Being at the same time a research field and a main topic for big tech companies (but not only), it ensures there is a continuous development of new tools, techniques, and studies to learn. In any case, I believe its’ mandatory to understand which is the best tool for specific a case to make the most of the available resources and minimize the development time of individual applications. If I must talk about a particular field of research, I believe that Transformers represent, but above all will represent soon the Next Big Thing also in our computer vision field.
Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Deep Learning World.
A: During my talk, I will talk about how deep learning techniques inside COPAN are changing the work of microbiological laboratories by speeding up and increasing their quality. Possibilities we generate are not only linked to a reduction in costs or an increase in productivity (quantitatively and qualitatively) but also an improvement in the patients’ experience to obtain accuratest results in a shorter period. Even more than techniques implemented, I believe that the real deal of our products is the decades-long microbiological knowledge combined with a huge dataset of clinical cases on which to apply any machine learning approach and idea.
Don’t miss Giovanni’s presentation, Deep Learning of Microbiological Analysis inside Full Laboratory Automations, at DLW on Tuesday, May 25, 2021 from 11:55 AM to 12:15 PM. Click here to register for attendance.
By: Luba Gloukhova, Conference Chair, Deep Learning World