Skip to main content

CAD WALK: Enabling Computer Aided Diagnosis of Foot Pathologies through the use of Metric Learning

Periodic Reporting for period 1 - CAD WALK (CAD WALK: Enabling Computer Aided Diagnosis of Foot Pathologies through the use of Metric Learning)

Reporting period: 2017-10-01 to 2019-09-30

It is estimated that anywhere between 17-41% of the general population experience foot pain and, in roughly half of these cases, the foot pain is disabling. Nevertheless, the majority of those with foot pain do not seek professional treatment. These statistics have been interpreted in the research field as a need for improving foot diagnosis techniques.

One tool that has shown potential in diagnosing foot problems is dynamic plantar pressure imaging. This technique measures, across time, the pressures that the bottom (plantar) surface of a person’s foot makes with the ground. From these plantar pressure videos, a clinician will attempt to identify pressure abnormalities and develop a personalized treatment strategy. Unfortunately, plantar pressure imaging creates large, dynamic, datasets with tens of thousands of sampled pressure values. These datasets cannot be easily analyzed and interpreted by the human brain. As a result, clinicians typically base their diagnoses on visual inspection of a small amount of this pressure data. By examining only a subset of the pressure video, it is possible that potentially valuable diagnostic information may be ignored.

We believe that a computer-aided diagnosis (CAD) system could efficiently assist clinicians in the analysis of full plantar pressure videos, thereby making sure that the diagnostic potential of plantar pressure imaging is fully exploited. To test this idea, we began this CAD WALK project with the goal of developing a CAD system for plantar pressure videos within a two-year time frame.

The CAD WALK system follows a basic principle: given an individual’s plantar pressure video, we will compare that video to one taken from the individuals “healthy identical twin”. If any differences in plantar pressures are found, then they must be related to the foot problem the individual is experiencing and should therefore be highlighted. Unfortunately for us, most people don’t have an identical twin, so we require a technique that will allow us to estimate their measurement.

Overall, the aims of the CAD WALK project can be listed as (a) the estimation of healthy plantar pressure videos from an individual’s demographic characteristics, (b) the use of these estimates in a CAD system to highlight plantar pressures that are abnormal, and (c) the creation of a commercial product out of this CAD system.
The creation of the CAD WALK system required the completion of the four following tasks.

1. Data Collection

In collaboration with Sint Maartenskliniek (Nijmegen, Netherlands), we collected 2640 plantar pressure videos from healthy individuals as well as 1011 plantar pressure videos from patients with foot deformities. The former group would be used to estimate the plantar pressures for the “healthy identical twin”, while the latter group would be used to test our CAD WALK system’s ability to identify abnormal plantar pressures.

2. Prototype Development

An initial prototype of the CAD WALK system was developed based on the concepts of metric learning and statistical parametric mapping. The prototype would use an individual’s age, gender, weight, height, and shoe size to estimate the range of plantar pressure videos possible for the “healthy identical twin”. That individual’s plantar pressures are then compared to this range and any pressures outside this healthy range are highlighted for the clinician.

3. Clinical Testing

The CAD WALK system was tested on a variety of clinical plantar pressure videos. From these tests, we were able to show that (a) previous plantar pressure analysis techniques examined only somewhere between 1-79% of the available information, (b) that by examining the full plantar pressure videos, CAD WALK is able to identify plantar pressure abnormalities that previous techniques could not capture, and (c) that patients with the same condition can exhibit different plantar pressure abnormality patterns, suggesting that different treatments may be identifiable based on CAD WALK’s results.

4. Product Development

Once the CAD WALK system showed clinical success, we refined the prototype software until it was easy to use by clinicians and operated in a short amount of time. Care was also taken to ensure that CAD WALK would run on multiple computers and agreed with industry regulations. Finally, we began to negotiate with companies about converting the CAD WALK system into a commercial product.

Overall, the project produces two large datasets of plantar pressure videos, two scientific articles, three international conference presentations, five workshop presentations, and a ready-to-commercialize prototype.
From a technological perspective, the CAD WALK project introduced new algorithms for the analysis of full plantar pressure videos. The project also refined these algorithms so that they can be used by clinicians who may not have a strong technological background.

The introduction of these algorithms was able to confirm that previous plantar pressure analysis techniques were not making full use of the information present in plantar pressure video. We were also able to show that foot complaints have larger and more diverse effects on plantar pressures than what had been previously reported. These results indicate that CAD WALK may enable more detailed diagnoses of foot complaints than previously observed.

Looking long term, the contributions made by the CAD WALK project have the ability to characterize foot complaints more thoroughly, allowing us to get a better understanding of how to treat such complaints. As the treatment of foot complaints improve, we suspect that people will be more inclined to seek professional treatment for foot problems, and as a result, the prevalence of debilitating foot pain will decrease.
Flow chart of the CAD WALK Project