Presentation by Joanne Grau, Head of Digitalization Thought-Leadership at Siemens Healthineers, Oct 31, 2022, for mHealth Israel- "AI and the Future of Healthcare". Three sections- Workforce Productivity, Precision Therapy and Digital Twin.
HOOD05162003252381
(re-use in parts HOOD05162003241203)
Our unique capabilities:
Patient Twinning –personalization of diagnosis and therapy selection
Precision Therapy – Intelligent and image guided treatment for the most prevalent diseases
Digital, Data and AI – advance providers’ operation with tech-enabled services
Today’s artificial intelligence is available to caregivers worldwide in commercially available solutions.
These solutions are primarily solutions to individual users or departments, not being connected to each other.
Already at this stage, these solutions can provide significant value to healthcare providers today:
Greater productivity of the individual, by automation of predictive tasks and avoidance of errors and the effort involved in their compensation
Consistent results, across patients, caregivers and institutions
Let me share some examples with you and for these examples I will take you into Radiology
Here an example for CT lung scans.
The lungs of a patient who has suffered COVID 19.
AI helps here to measure the impact of the COVID-19 disease on the lung
It automatically detects and quantifies the extent of total COVID-19 abnormalities based on 3D segmentation of lesions, lungs, and lobes.
During the infection – decide for admission to hospital and potentially ICU
Has been used more than 350.000 times
Other than most human operators, AI scans all chest CT datasets for other incidental findings
Smoking for example increases the likelihood of various severe diseases, lung cancer being of them
AI automatically searches for abnormalities in chest CT datasets and can for example identify coronary calcifications or dilatation of the thoracic aorta
In the event of incidental findings, the operator is notified and the corresponding report automatically pre-populated – which is a great opportunity to provide consistent, high-quality diagnosis to each patient
Another example includes MRI of the prostate to diagnose prostate cancer, which is one of the growing areas in oncology
Prostate MRI is an expert task, and the growing patient numbers put additional pressure on senior radiologists
Artificial intelligence reduces the workload of experienced radiologists and at the same time provides guidance to less experienced radiologists to successfully perform this expert task
AI automatically segments the prostate and detects lesions. These lesions are automatically quantified and the PI-RADS score for severity assessment is calculated.
The operator has the opportunity to correct the findings if needed, and AI concludes the task by populating a standardized report.
A recently published study compared the findings from experienced and less experienced radiologists. The study concluded, that less experienced radiologists can provide the same level of quality as experienced radiologists, when accompanied by Artificial Intelligence.
https://link.springer.com/article/10.1007/s00330-022-08978-y
This is an example from the emergency department, where time is critical
One of the promising applications is running artificial intelligence on the CT scanner.
In a stroke patient for example, AI can automatically detect a suspected intracranial hemorrhage and alert caregivers.
This helps to prioritize these critical cases, where minutes can be decisive for the future of the patient’s life.
These case can be read by radiologists in the control-room, or in their usual reading environment
This concludes today’s applications and I now want to look into the near future
The next era of artificial intelligence, which is subject to current research, will greatly help to deliver more precise therapy
By aggregating data along clinical pathways, Artificial Intelligence will be able to guide treatment decisions in the interest of achieving the best possible outcome and minimizing side effects
We think AI can help with this and have been collaborating with multiple healthcare institutions to try and create a predictive algorithm that could be used to help identify the potential risk a COVID-19 positive patient may have of sever illness from uncontrolled inflammation (Cytokine Storm)
We’ve worked with leading healthcare institutions from Atlanta, Houston, NY and Madrid to collect and analyzer de-identified patient data and then Created AI-based algorithm to predict likelihood of:
acute respiratory failure
end organ failure
30 day mortality
We used the data from the cohort to train the algorithms with a combined data source of ~14,500 COVID-19 diagnosed patient cases
Our goal was to find the right balance of including the minimal amount of IVD test results needed to yield accurate prediction scores and ended up including Age + 9 lab different biomarkers generated within the first 3 days of hospitalization (D-Dimer, LDH, Lymph %, Eos %, CREAT, CRP, FER, INR, Troponin-I)
We are now have a couple healthcare institutions performing Investigational Use Only evaluations to assess the potential clinical utility and benefit of the algorithm
You can see here the current prototype being used that includes test result entry on the left side with the overall patient severity score and 3 individual risk scores for acute respiratory failure, end organ failure and 30 day mortality displayed on the top.
We feel this is an important example of how AI can and will be more commonly used in the future to help create tools to predict outcomes and aid physicians in making clinical decisions.
Let’s look at an example on lung cancer treatment
By taking as an input the treatment parameters, tumor information, and outcome data, we can learn the fingerprint to differentiate responder and non-responder groups.
We have demonstrated that the local failure rate to radiation therapy can be reduced by 45% in the favorable sub-group and that AI can also help modulate therapy.
This way, AI helps to precisely prescribe the required dose for radiation treatment for the individual patient.
Now I want to discuss and example, on what is possible when connecting information across the patient pathway – on the example of lung cancer
Based on the prediction of the likelihood of patients developing cancer (colorectal, liver or lung) over the next 12 months, which is subject to current research. AI can notify the physician and the patient of the need for diagnosis, for example by chest x-ray or chest-CT.
As we saw earlier, detection, highlighting and quantification of findings is commercially available today.
AI also helps today in automatic contouring of organs at risk, which is required for radiation therapy.
The next step in the patient pathway is treatment planning. AI-based treatment planning, which has the potential to greatly help increase the productivity of caregivers and to account for changing conditions of the patient during the radiation treatment cycle, e.g. weight loss. The latter will of course contribute to a more precise treatment to the individual patient.
Looking further into the future, our vision on how to use data and Artificial Intelligence, is to create a digital twin.
A digital twin of the patient – and when we look at prevention this also includes healthy individuals.
Imaging, what if we could create a digital twin of the patient’s heart?
Let us move forward and bring into the equation the physiology, the mechanistic approach of explaining how things work, in addition to statistics.
Also include guidance on therapy decision, is it effective or not, predict will therapy be effective
What is one of the advantages of having a virtual patient-specific heart?
Let’s look at different applications
Different arrhythmia developments can be simulated on the patient. This might allow to test whether the patient will develop an arrhythmia thanks to biomarkers of the model. Moreover, once the patient has the arrhythmia, different treatment options can be simulated to have a suggestion of the optimal ablation procedure or the optimal placement or timing of a pacemaker.
Imagine – at some point in the future our data is being integrated from birth through a lifelong, physiological model that is updated with each scan and exam. And specifically designed Neural Networks are analyzing the data continuously. As a result, we don’t talk only about disease care, but we focus on health care, person centric prevention, When disease happens we have the right tools to establish the correct diagnosis and select the best treatment.
For this vision many things have to evolve, specifically on the AI technology.
Looking at the rapid development of computational power – in supercomputers and also at point of care – I’m optimistic that we will see the digital twin of the patient in the not-so-distant future.
Person includes both – patients and healthy people
Workforce productivity
Efficiency – Automation of repetitive tasks
Consistency – avoidance of errors helping to reduce effort on correction and consistent high quality