3D Printing + Sensors + AI= Shoe Insoles 👌

PHEET — an innovative solution for custom foot orthotics

Charles Fried
May 18, 2017 · 6 min read

This article was originally written for 3DMedNet

I believe we are now a point of convergence where additive manufacturing (also known as 3D printing) is becoming a cost effective solution to deliver mass customized products. The nature of this technology lacks economy of scale but allows for a high level of customization making it an ideal mean of production to respond to anatomical variances on an individual basis.

With the underlying knowledge of this technology I was listening to my brother speak about his seemingly endless search for a solution to cure his foot problem which kept him away from his passion — running. His search started from a simple off-the-shelf (OTS) product, then onto a semi-custom product (SCP) which was thermo-formed and finally onto a fully custom orthotic (CO) solution which was conceived by a Podiatrist. The former two actually worsened the pain whilst the latter made an improvement.

There were two key points which struck me in this process; the first was the gap of price and quality between the SCP and CO solutions and the second was how lengthy (2 weeks) and invasive the CO production process was. Upon this realization, and equipped with some software and computer aided design (CAD) skills, I embarked on my journey to re-think how orthotics are delivered, by leveraging technological tools.

My research started by reading a number of papers and getting a basic understanding of biomechanics and the practice. One that particularly stood out was research by Ingrid Knippels (MOBILAB; Belgium), who carried out an extensive study with 9 experts and 67 subjects to understand the extent in which foot experts differ in gait analysis [1]. Some features were more prone to disagreement than others, but at worst, the experts could not agree on whether the calcaneus in relaxed stance had a valgus deformity or was normal. In layman’s terms, this is a condition whereby the heel is rotated outwards.

This level of subjectivity led me to artificial intelligence (AI). Despite my humble experience with artificial neural networks (ANNs), my intuition had me believe that if there was a way to obtain some data which had a proven correlation with pathologies, the ANN could be trained into diagnostic system.

ANNs learn through a process called back-propagation which involves feeding the errors backwards through the network in order to adjust the connection (synapses) between each neuron. From the data set used to train the network we know whether the patient has a pathology or not and this can be used to train the network accordingly. Wafai et al did just that, by collecting the plantar pressure of 47 subjects to train an ANN [2]. Their model achieved a classification accuracy of 87–100%.

With a basic understanding of the foot’s anatomy and biomechanics I interviewed a number of podiatrists and asked them to describe the current process. What struck me here was that although they were all equipped with advanced technological tools such as pressure plates and video analysis, unfortunately none of those had a direct input into the final insole. Rather, the podiatrist filled in a standardized form to communicate the requirements of the insole to the orthotist who would manufacture it. With this process, it is very common to require post-manufacturing alterations, necessitating the patient come back a number of times to arrive at an adequate solution.

With Waifai’s work in mind, which analyzed static pressure data, I aimed to build a device that could collect dynamic data during periods of exercise. This would solve the problem of patients unwillingly changing the way they walk whilst under observation, as research has shown. It would also produce data more consistent with the final use of the insole. This would make a more successful product, as pressure has a direct correlation with comfort.

The device produced was an insole with 16 sensors per foot, and a micro-controller that reads and stores the values 60 times per second, embedded within the insole.

The software utilized can be divided into two parts: the analytics and the 3D modelling. Both of these are linked using a configuration file. This is a key point of the project, as by doing so we created a close coupling between the pressure data and the geometrical data used for manufacture. Ultimately, this give us a much higher degree of control over the behaviour of our insole.

Live Replay & Config User Interface (Prototype)

One of the analytical methods used involved analyzing the gait line. We obtain this data by plotting a line of the average pressure distribution through time. Now, if we considered the anatomy of the foot to be a three pointed arch, then the optimal gait line (below figure, grey) resembles a smooth S-like shape, which is drawn as the pressure shifts from each point in a forward and sequential manner.

Illustration of gait line alignment

We can also quantify how much the actual average gait line deviates from our ideal (above figure, red line). We do so by breaking down the line at equal points on the y-axis and measure their distance on the x-axis. From there, we can attempt to correct the disparity by assigning a harder material to the opposite side to “push” the gait line closer to our ideal.

Once we have a diagnosis, the behavior of the insole can be controlled by altering its geometry and by modifying the material allocation. The idea of producing an object which has varying properties throughout to respond to local requirements was largely inspired by the work of Neri Oxman (Massachusetts Institute of Technology; MA, USA) who coined the term Variable-Property-Rapid-Prototyping.

The biomimetic pattern is derived from an algorithm called Reaction-Diffusion which allows us to control friction by varying the density locally, in turn increasing the contact area. Lastly, by using a multi-material 3D printer, we can assign different hardness of plastic to each segment of the insole based on the local requirements.

Final Product- PHEET

The final system is able to produce an insole based on 15 minutes of exercise, a 3D scan of the foot, and, at this stage, the input of a trained professional. It can then be manufactured locally using the closest 3D printer and delivered to the patient the following day; a substantial improvement on the current 1–2 week delivery time.

This project was undertaken in 4 months and still requires some further research and development before becoming a viable solution. I’m currently looking for partners who share the vision of a future of mass-customized medical products who can help me explore the potential of this system.

Another potential application of this technology addresses the diabetic foot. Foot problems are a complication of diabetes which in the UK alone causes 135 amputations every week. With adequate foot care, up to 80% of these can be avoided [3]. In most cases, diabetic foot will start as ulcers, where orthopedic treatment can be effective. It works by taking the pressure away from the paint point and redistributes it to the surrounding areas. This can be achieved utilizing the PHEET device.

I hope you’ve found this article somewhat interesting and informative; I feel we’re at a very exciting period of exponential technological growth. I think amazing innovations will happen as a result of the synergy between AI and digital manufacturing especially within the medical industry. If you have any questions, please get in touch.


  1. Knippels I, Saey T, Van den Herrewegen I, Broeckx M, Cuppens K, Peeraer L. Comparison of biomechanical foot analyses between nine Flemish foot-experts. J. Foot Ankle Res. 7(Suppl 1): A45. (2014)
  2. Wafai L, Zayegh A, Woulfe J, Begg R. Automated classification of plantar pressure asymmetry during pathological gait using artificial neural network. 2nd Middle East Conference on Biomedical Engineering (Doha). pp 220–223 (2014)
  3. Diabetes UK. More than 135 amputations every week. Public Health England Data (2015)

Charles Fried

Written by

Tech Lead @ Additive Flow & Founder @BlockSmith Capital. Keen Investor. Talk@charlesfried.com