Machine Learning Applications for Medical Imaging Devices

Mohammed Guller

In modern healthcare, medical imaging devices such as X-ray, Computed Tomography(CT), Ultrasound, and Magnetic Resonance Imaging (MRI) devices play a critical role. These devices allow healthcare professionals to examine the patient and determine the root cause of their symptoms. Imaging devices allow healthcare providers to develop the right treatment plan for their patients.

These imaging devices are expensive, and the average cost is around a million dollars. In addition, these devices have parts that frequently break and are expensive to replace. Some of these parts cost hundreds of thousands of dollars. Therefore, not many healthcare facilities can afford to keep spare parts. Generally, a replacement is ordered when a part fails. Sometimes, it takes up to 2-3 days for a replacement part to arrive. Until then an imaging device cannot be used. Thus, an expensive asset stays unutilized. In addition, a facility may have to send a patient to another facility if this happens. As a result, it may lose revenue and get unhappy customers. Thus, reactively replacing a failed part has many undesirable consequences. A facility has to bear not only the cost of a replacement part but also lost or deferred revenue and dissatisfied patients.

Another issue that health care facilities and medical imaging device makers have to deal with is tracking the sensors embedded in these devices. A typical device may have hundreds of sensors monitoring different things such as tube temperature, fan speed, air temperature, water flow, and others. It is difficult for a human to keep track of which sensor readings are normal and which ones are not. Usually, people use a rule of thumb to define thresholds for flagging a sensor reading as normal or abnormal. However, these manually-defined thresholds may misclassify normal values as abnormal and vice-versa. Generally, an abnormal value is an indicator of a problem. Missing these signals means missing an opportunity to take corrective action in time. As a result, a problem may go unnoticed and result in bigger problems that are more expensive to fix.

The third topic that I would like to discuss is capacity planning. As population grows and more people get healthcare coverage, the demand for medical imaging continues to grow. A healthcare facility needs to buy more medical imaging devices as demand grows. However, capacity planning is a challenging task. On one hand, purchasing a new device too early is wasted CapEx on an under-utilized asset. On the other hand, not having enough devices to satisfy demand results in lost revenue and customer loss. Therefore, it is important for a healthcare facility to track usage of its devices and be able to forecast future demand.

All the challenges described above can be effectively addressed with machine learning. It can be used for predicting part failures and proactively replacing parts to minimize unplanned downtime of an expensive asset. Similarly, it can be used to detect anomalous sensor readings and fix problems before they snowball into more expensive-to-fix problems. It can also be used for capacity planning. It provides tools for forecasting future usage based on current usage and other factors.

Glassbeam’s machine learning (ML) based solutions addresses all the above challenges. We have the technology and the required data for building effective ML models. I will be covering these in more detail in future blogs. Watch this space. Do signup for updates.