• Machine Learning
  • Robots could be good for your health

    In his book The Rise of the Robots, published by One World, Martin Ford proposes social and economic scenarios for robots that are good for output, but not so good for people. He sees significant upheaval and displacement from employment across a wide range of commercial and industrial activities and across middle and low income families. The drop in income, so spending power, will degrade economies.

    Simultaneously, robots aren’t paid and don’t spend money. He sees this as exacerbating the social and economic impact.

    Healthcare’s the activity that’s different. He sees the robots marching into healthcare that’s already over-stretched as needs and demands continuously outstrip supply. Four roles are crucial:

    Artificial intelligence in medicineHospital and pharmacy roboticsRobots that care for the elderlyUnleashing the power of data. 

    For low and middle income countries and health systems, sustained investment in robots could be part of the solution. They can improve healthcare professionals’ productivity and help to meet demand.

    They should find a place in Africa’s eHealth strategies. Small scale investment will lay out a trajectory for the future.

  • AI and machine learning need data storage resources

    Many things come in bundles. Amit Ray, author of Mindfulness Meditation for Corporate Leadership and Management says “As more and more artificial intelligence is entering into the world, more and more emotional intelligence must enter into leadership.” It’s not enough. A report by Source Media, sponsored by Pure Storage says powerful, advanced computing and storage capacity and capabilities are needed too.

    It recognises AI’s “vast” potential. Currently, some radiology departments use it effectively to improve workloads. Progress across other clinical activities depends on extra computing and storage power for two activities, training and clinical use.

    When researchers deliver AI and machine learning techniques to clinical practice and healthcare, solutions need huge amounts of data for training models, including labelling data. It’s especially important for neural networks. These are hardware and software patterned on the way neurons work in human brains. They’re deep learning technologies often focusing on solving complex signal processing or pattern recognition problems.

    If storage’s inadequate, it can’t keep up with the workload. The result’s diminished AI. Healthcare’s typical eHealth investment model’s to buy enough computing storage infrastructure as a minimum requirement, then expand it a few years after it’s clogged up. Eventually, it’s replaced with modern solutions after a period of obsolescence.

    This doesn’t fit AI and machine learning. It has to match the computer power and storage capacity needed as AI and machine learning expands. Developers and healthcare organisations can then move beyond exploring AI’s potential and bring into full use. The, patients benefit.

    While assembling the resources needed for AI and machine learning’s challenging for Africa’ health systems, the infrastructure requirements add to the constraints. Before venturing into the AI space, it’s essential to contemplate and deal with the whole resource requirements and their affordability. 
  • Computer aided detection for TB (CAD4TB) installed across Ghana

    A target of sustainable development goals (SDGs) is to end tuberculosis (TB) globally by 2030.  Effective prevention, detection and treatment is necessary to achieve this goal.  Ghana is in the global high burden list for TB, but is tackling this burden using eHealth innovations. 

    In collaboration with Delft Imaging Systems, they have successfully installed 51 X-ray systems in facilities, containers and TB screening mobile clinics across the country.  These mobile X-ray systems are self sustainable, employing solar technology to power them, even in the remotest of locations.  All X-ray systems have been equipped with computer aided detection (CAD4TB) software that makes use of machine learning to detect TB in X-rays.  Additionally, tele-radiology technology is used to interlink all images to a central platform that allows healthcare providers across connected facilities and units to access images.

    The innovation allows healthcare providers to screen up to 200 images per day.  When the images reveal a high CAD4TB score, patients are referred for the standard and more expensive GeneExpert tests.  This makes detecting TB in poorer communities very effective. 

    It is eHealth innovations like this that will strengthen health systems in Ghana and other African countries, while still being conservative of the constrained health budgets in Africa.

  • A smart watch can detect epilepsy

    Epilepsy is a leading serious neurological condition worldwide.  It has particularly significant physical, economic and social consequences.  Recognising the need for an intervention, Empatica Inc. has developed a smart watch to detect seizures in epileptic patients.  They’re calling it Embrace.

    Embrace uses machine learning algorithms to monitor and detect different seizure types, including grand mal or generalised tonic-clonic seizures. Electrodermal Activity (EDA)* sensors in the watch are used to measure multiple indicators of a seizure. 

    It’s also accompanied by an app that will send an alert, via text message, to a healthcare provider or caregiver once a seizure is detected.  Additionally, the app serves as an electronic seizure diary and health record for the user.

    During a clinical study involving 135 epileptic patients, Embrace’s algorithm was shown to detect 100% of the seizures, including the 40% of silent seizures that were unreported in patient clinical diaries.  Following this, the smart watch has received FDA approval as well as approval in Europe as a medical device for epileptic monitoring.

    Embrace’s high sensitivity is revolutionising seizure reporting.  It serves as a much awaited alternative to wearing an EEG, that is automated, and isn’t bulky or cumbersome to wear.

    *signals used to quantify physiological changes in the sympathetic nervous system 

  • Voice recognition reduces Tanzania's patient waiting times

    Patients at the Muhimbili National Hospital in Dar es Salaam no longer have to endure long waiting times for their radiology results.  This is thanks to a new technology installation in the department.  Voice recognition or speech recognition technology is now being used to encode doctors notes on patients so that they can easily be transferred to the radiology department. 

    With this new technology, Tanzanian medical professionals are able to dictate into their computers, in the normal course of speaking and have the speech engine recognise what the clinician wants, and then apply the commands or structured words, respectively, to obtain a radiology report for a patient.  There has been some concern around the effect of speech accents on the technology, but this has posed no problems since implementing it at the hospital.  

    The speech engine is also capable of showing the cardiology report template populated with the name of the patient and other demographic data. By dictating the cardiology report narrative, the computer recognises the narrative context and intent and condenses a complete, correct, and structured document.

    This translates to shorter waiting times for patients, greater operational efficiency within the hospital and reduced workload on medical staff who are required to take notes of patient examinations and consultations.  The technology, which uses natural language processing, is constantly learning speech behaviour through repetitive exposure to terms and complex algorithms that organise speech patterns into recognisable behaviour. 

    This bold technology implementation in Tanzania could be a useful pilot for overburdened health care systems in Africa hoping to achieve the same benefits.