Are you actively implementing or have plans to execute an AI strategy

0
137
Imaging

By: Dr. Amit Kharat MBBS, DMRD, DNB, Ph.D., FICR (Radiology) Co-Founder, DeepTek.ai

The radiology industry needs to have a fresh perspective on the legacy systems being used in the radiology space and completely rethink how AI can enhance the experience for the stakeholders. In the radiology and imaging space, the stakeholders are the patient, imaging experts (radiologists), and referring physicians. The idea is to ensure how AI can enhance the experience for all concerned in multiples of x.

AI Versus Smart Solutions:
AI is a blanket term and is not appropriate as it does not do justice to its actual workings. The appropriate word must be smart solutions though it sounds less glamorous. A smart solution can use AI or simple rule-based engines and decision matrices to enhance the user experience and produce decent outputs. These smart solutions may also use machine learning and its subsets, such as deep learning to complement. All in all, this makes a complete end-to-end platform for radiology. Such smart radiology solutions principally can make things efficient, and economical and also be a strong enabler.

Point of Impact:
Smart Solutions like Augmento by DeepTek can be deployed in radiology departments for appointment scheduling, the process of patient scanning, scan result interpretation (prescreening and triage), workflow augmentation, and imaging result delivery through notifications. Such solutions will significantly increase the value radiology professionals can provide to their patients.

Deployment Strategy:
Deploying AI for workflow and image analysis is complex. This also needs to be smoothly integrated into the radiology workflow. The aim of the activity is to reduce bottlenecks and workflow congestion. If AI has increased the number of clicks or added an additional dashboard, then it’s not going to be easily accepted in the workflows. In fact, how we can minimize the clicks, for example, a simple few click reporting for normal studies. Currently, a lot of time and effort of floor managers is wasted in rationalizing resources and in the assignment of studies to experts in the radiology department, which ultimately results in valuable time getting wasted. These tasks can easily be automated. Workflow automation and combining it with radiology medical image analysis, integrating it with structured report generation could result in substantial time savings in the radiology department. Executing this in the right way is the key to obtaining success and streamlining the work process within the radiology department.

AI strategy and use cases:
Bringing automation to radiology and medical image analysis requires us to dive deeper into the use cases for AI in the radiology department. Let’s take a simple example of an X-ray chest and see the variables involved. It starts with a seemingly simple task of finding normal versus abnormal findings and segregating these x rays. Then it can move further to a more complex task of detection of particular disease condition like TB and Covid 19 pattern or other diseases; it can move further be broken down to more complex problem solving such as detection of additional lung pathologies like pleural effusion, consolidation, mass, nodule, cardiomegaly (this is yet an oversimplification). The AI models will also segregate and detect/exclude skeletal (bone) abnormality in the x-ray chest, such as scoliosis, fracture and as we continue to dive deeper, we can go to more depth such as understanding nuances like the quality of the chest radiology, motion blur and rotation of the x-rays all detrimental to image analysis and interpretation. Such radiographs with quality can be caught early on and can be flagged.

Creating a well-organized chest suite for the x-ray itself requires deep knowledge and high-quality annotations. Of course, we should not forget that the same pathology can have multiple differentials, and we need to be more careful while annotating and creating a list of differential diagnoses possible. Therefore the annotation exercise is intertwined with the clinical outcomes captured and final diagnostic details available for the given study during the process of data annotation. This process can ensure robust algorithms.

Finally, all AI Models will essentially, at some point in time, need strong feedback and self-learning through actual modifications being brought around by radiologists as this feedback can establish the robustness of the algorithm over a length of time. However, running such complex chest solutions have cost implications. Hence, implementations must be carefully planned and executed by weaving the Ai deployments into the long-term strategy of the organization of deploying AI in the vision and goals for the next 5 and 10 years of the radiology department.

In conclusion:
Every organ and every imaging modality in medical imaging has its own unique set of challenges and opportunities to better healthcare. While executing an AI strategy, one needs to ascertain using a simple checklist.

  1. Whether the planned solution rollout is improving workflows.
  2. It is empowering experts.
  3. Does it ensure significant saving in terms of time, costs, and minimizing errors? Healthcare imaging automation, therefore, is not a simple exercise of a custom model deployment but a much more complex activity to ensure things work and enhance the experience for users.

Besides India, many countries in Africa and Southeast Asia have also adopted AI solutions to address the shortage of radiologists. AI is not only making TB diagnosis accessible to people who were previously deprived of it, but by reducing radiologist effort required in reading X-rays, it is lowering healthcare costs and improving reporting times.

LEAVE A REPLY

Please enter your comment!
Please enter your name here