“AI 4 the health value chain”: update on AI solutions for medicine
Following the last virtual event “AI 4 the health value chain” of the project Ai4diag with the participation of five of the most innovative health tech Smes on the market, allowed us to have updates on the state of the art of artificial intelligence applications in biomedical and medical diagnostics.
On the other hand, the systematic sharing of updates on emerging technologies is a natural practice in a company like JSB Solutions that has two founding values of its mission in experimentation and innovation. Cross fertilisation with methodologies adjacent to those on which we are focused (blockchain in the first place) opens the way to new ideas and the improvement of our services for the pharma and life science world.
Big data, natural language processing, machine learning, neural networks and deep learning are the different faces of a computational methodology that is having a huge impact on the entire value chain in the medical and health sector.
Their development will be crucial in the processes ranging from basic research to the production of drugs, from regulatory affairs to diagnostics; it will increasingly influence trials on patients in clinical trials as in predictive medicine; in the therapeutic choices up to the follow-up of diseases and cures.
Machine learning and new challenges
In the last years, the development of machine learning techniques has made it possible to address for the first time complex problems that could not be solved with traditional programming methodologies, based on lists of rules formulated ad hoc, based on a deterministic and probabilistic approach, not designed to react flexibly to exceptions.
With machine learning the approach becomes heuristic: in solving problems the algorithm no longer follows a predetermined path, but acquires the tools to learn on the basis of empirical data, so as to mature experience and generate knowledge.
The main challenges of the present are to search for relationships between huge sets of unstructured data to find correlations useful to determine therapeutic choices; to refine predictive analysis for the development of personalized medicine; facilitating patient monitoring and the delivery of large-scale precision medicine according to validated standards.
In this particular period artificial intelligence is having for example wide application in predictive analysis and diagnostics of patients Covid-19, a bit around the world.
Apart from pandemic, in addition to predictive analytics, AI also finds its most recent fields of application in genomics, to model genetic sequences, and in telemedicine (teleassistance and tele-rehabilitation, with the help of wearable sensors).
State of the art AI solutions for diagnostics and non-diagnostics: some examples
Beyond this, there was no lack of practical examples of speakers at the event on AI solutions in the most diverse fields of medicine, based mostly on image processing and natural language processing, while there are already many biomedical applications of artificial intelligence validated and approved by FDA. Among the most interesting cases reported in the speech:
- “Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs” – A study on the development and validation of deep learning for the identification of diabetic retinopathy in photographs of the retinal fund, comparing the performance of the algorithm with the manual evaluation of ophthalmologists.
- “Deep learning algorithm predicts diabetic retinopathy progression in individual patients” – As the deep learning algorithm provides the progression of diabetic retinopathy in individual patients.
- “Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning”– As neural networks have been able to predict multiple cardiovascular risk factors based solely on images of the retinal background, using anatomical features, such as the optical disc or blood vessels, to generate each prediction.
- “A study of deep learning approaches for medication and adverse drug event extraction from clinical text” – A study on deep learning approaches for extracting drug data and associated adverse events from clinical records.
- “Assessment of deep natural language processing in ascertaining oncologic outcomes from radiology reports” –
A study on the evaluation of deep natural language processing in the assessment of oncological outcomes from radiological reports, able to detect the most important endpoints of clinical studies, such as response to therapy and disease progression, with statistically significant associations between oncological endpoints and global survival.
Much of this information without AI and Machine Learning algorithms would be inaccessible, as the data from which it is extracted is often recorded in electronic health records (EHR) only as free text or in unstructured format (photographs, ultrasounds, Tacs)