Recognizing voices in the trickiest terrain: The battlefield
It's not here yet, but the United States military has plans to perfect technology that would enable combat medics to dictate treatment notes for casualties on the battlefield – and have those end up as structured data and narrative information in fighters' electronic medical records.
The military has long struggled to provide end-to-end electronic documentation of casualties' treatments from the battlefield to battalion aid stations and so on through multiple levels of care until the fighter is evacuated to a hospital.
Findings, procedures and medications provided by a battlefield medic, however, are important to providers at the next level of care for the sake of the efficiency and effectiveness of the follow-up treatment. The military also wants to capture this information for the sake of researching how to treat severe trauma at the point of injury.
"We want to get these records into structured text so that we can do data mining," said Lt. Col. William Geesey, product manager at the Army's Medical Communications for Combat Casualty Care (MC4) organization. "Information gathered at the point of injury can inform research on better methods and technologies for treating wounded soldiers. That can increase survival rates."
Shedding paper, embracing electronic forms
Getting battlefield information up through the next levels of care is easier said than done. Originally, the Army devised a paper form known as a field medical card for medics. This was to physically accompany the casualty up through to the next level and was to be eventually entered into the electronic record. For a variety of reasons, that card often didn't show up with the patient.
Next, MC4 deployed an electronic version of the same form which medics accessed on handheld devices or laptops. Here, again, medics found it difficult to punch in data on a touchscreen while treating the patient.
These difficulties spawned the idea of developing tools that would allow medics to dictate notes in real-time, eventually having voice recognition software record those notes. Later, they could be used for populating fields on the form as well as providing additional narrative information. It's not exactly an original idea – the technologies which enable this type of process already exist in the civilian world – but there are a number of challenges with adapting these to the battlefield, and research is ongoing to accomplish those ends.
In the civilian healthcare world, physicians ranging from emergency doctors to surgeons, radiologists and pathologists are able to dictate notes, which are then automatically transcribed into text. "The idea is to go from voice to data and information," said Monika Woszczyna, head of the speech technology research department at M*Modal, a Pittsburgh-based company which provides speech understanding technology to EHR providers. "The technology is able to understand words and medical concepts. The output is electronic text which is processable, analyzable and shareable."
In most cases, physician users of M*Modal technology dictate a report which is sent to a secure server until picked up by a transcriptionist. The transcriptionist makes corrections to the software-generated text because, Woszczyna said, "recognition errors are unavoidable. The question is how many you get. "Those in the loop are also important because "from what the transcriptionist fixes, the system learns at all levels, including differences in particular voices and topics associated with different medical domains. The feedback loop reduces error rates significantly."
But there may be situations, and the battlefield scenario is likely one of them, where providers can't wait for the transcription, which can take several hours, to unfold. M*Modal's technologies allow text to be generated in real or near-real time while preserving the audio recording for further reference.