The Future of Physical Therapy (PT) and Artificial Intelligence (AI)
What is AI?
AI = Artificial Intelligence and it simply means a computer acting human like – making decisions, identifying objects, and having voice recognition. AI will seemingly effect every industry at varily levels. But it is still unclear how AI will directly impact physical therapy. Therefore, let’s take a close look at AI and PT together. Currently, there are two basic forms:
◦Machine Learning
◦Deep Learning
What is Machine Learning?
Machine learning or ML occurs when a code is entered into a computer to allow the computer to “train” itself. Instead of entering many codes, only one code is needed. Then the computer is fed millions of datasets. The computer will “learn” to Identify the data. For example, a computer is given the code to identify a cat. It is then shown millions of photos of cats and told that these are cats. After the computer “learns” to identify the cat images, it will be able to identify a cat photo without being told it’s a cat. It will inform you that it has identified the cat.
What is Deep Learning?
Deep learning is more complex and largely based on deep neural networks (DNN). Much like the brain, there are multiple pathways to get a solution. In this type of AI, the computer is fed layers of information and it will “learn” to identify items based on features. For example, a computer is shown pictures of mammals, birds, and reptiles. Next the computer is shown a lizard. The computer will then be able to identify the lizard as a reptile based on the scales, colors and characteristics.
What AI can currently do in PT::
1. Measure ROM with more accuracy
2. Give patient feedback about their home exercise program (HEP)
3. Roughly diagnosis pain as musculoskeletal disorders
What AI needs to be able to do:
1. Measure range of motion (ROM) and strength output at each joint
2. Measure gait and standing posture
3. Assess correct movement at each joint
4. Give real-time feedback to a patients about their movement
5. Develop exercise protocols for specific diagnoses
6. Analyze outcome data on patient recovery
7. Modify exercise protocol based on outcome scores
8. Diagnosis a general disorder
0. Identify “red flags” for medical emergencies
What AI cannot do:
1. Assess joint mobility and end feel of joints (arthrokinematics)
2. Apply soft tissue techniques such as myofascial release (MFR), soft tissue massage (STM), or use instrument assisted soft tissue mobilization (IASM)
3. Hug the patient and smile at them
Current state of PT and AI
1. PT is the top 5 most expensive medical service according to Medicare
2. Chronic pain affects millions of people in the US and many times can be treated with exercise, massage and mobilization of joints
3. Insurance companies are basing their reimbursement on standardized patient answered outcome scores (KOS,LEFS, NDI, etc)
4. Preliminary data shows that AI produces better patient outcomes
5. AI can look at kinematics but not arthrokinematics
Treatment of a Total knee Replacement (TKR):
The patient walks in the clinic with a rolling walker, knee swollen and not fully able to bend or extend the knee. The patient is also limited in balance, ambulation, stair climbing, sitting tolerance and gets very uncomfortable sleeping.
On the first outpatient session, the PT will assess the ROM, amount of edema, the strength deficits, limits in balance, and abnormalities of gait. The PT will then work on edema control with STM, educate the patient in elevation therapy with icing and check for any signs of infection. The PT will then show the patient how to complete ROM exercises and start the progress of strengthening through isometric exercises. If the patient can tolerate it, the PT will also start manual work on the joint by moving the patella and knee. The PT will work to sense the end feel of the joint with patient feedback. The PT will then educate the patient about regaining full extension to allow for full return of a normal gait. The PT will have the patient walk, possibly with a cane, to help the patient relearn how to heel strike, stride through and keep the knee extended in the stance phase of gait. At the end of the hour session, the patient may feel more ROM, less pain, and more confidence with ambulation
Enter AI…
Prior to coming to the first outpatient PT appointment, the patient was given his own AI device. Thus far, the AI has been able to take measurements of the patient’s ROM and edema and has started the patient on a ROM and isometric program. The AI device has also been able to assess the gait motion of the patient and has started to encourage the patient to strike his heel, straighten his knee and swing through his leg while using his walker. The PT has received this data, and can further assess the patient's strength, balance, and to start the stair climbing process. The PT will also apply joint mobilization techniques and edema control therapy to the patient while educating them on the importance of regaining full knee extension. Next the PT will show the patient how to properly squat and will add this to his HEP and integrate it into the AI program. The PT will reassess the gait pattern and if safe, will move the patient to a cane for ambulation. If there are no signs of infection, the edema is well managed, and ROM has improved since surgery, the patient will return to PT in 1 -2 weeks for a follow up session with the PT. In the meantime, the patient will continue with his AI device and continue with ROM, strengthening exercises and gait training. The AI device will communicate with the PT and send feedback about each daily session. If the PT notices a loss in ROM, limitations in gait, lack of strength or difficulty with balance, the patient will be asked to return to the PT clinic.
This could be the future of our rehabilitation world. Currently most AI devices are only able to use machine learning and wearables to guide patients through an exercise program.. But, soon, neural networks will be designed to not only assess correct movement, but will also be able to assess incorrect patterns, and adjust their output based on the feedback from the patient.
For more information, please visit www.saramikulsky.com
