SDG 3: Good Health and Well-being
2. Project Details
Company or Institution
Data Scientist Network Foundation (Data Science Nigeria)
General description of the AI solution
PrecribeWrite is an AI powered Optical Character Recognition (OCR) system that aims to mitigate medication errors through automated digitization of Handwritten Medical Prescriptions in Low Resourced Health Systems.
Data Science Nigeria
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
The core of this project explores the use of AI-enabled optical character recognition (OCR) technology trained on context-specific (prescription-related words) data to help mitigate prescription errors.
Medical diagnosis and prescription documentation are not digitalized in many parts of the emerging market. Hand-written prescription by medical personnel is a common trend, and this comes with a significant risk of medication error and wrong medication usage due to varying and illegible handwriting.
In emerging markets with low-resourced health care systems, there is a shortage of qualified medical personnel and most of the drug dispensing is done by informal healthcare providers. This practice comes with a considerable burden of medication error due to illegible hand-written prescriptions. This explains the rationale for an automated digitalization interface to translate all hand-written prescriptions into a digital format using an optical recognition system.
The goal of the project is to harness artificial intelligence-enabled optical character recognition (OCR) which is the electronic conversion of scanned images of hand-written text into machine encoded text on hand-written medical prescriptions to improve health delivery outcomes in low-income countries.
The project methodology is listed below.
1.) Detailed process documentation
2.) Data gathering and preprocessing
3.) Optical character recognition (OCR) model training and iteration
4.) AI enabled model deployment with API end points to serve applications
This project is currently at stage 3 of the methodology where different deep learning algorithms and frameworks such as TensorFlow, Pytorch and Keras are being tested to select the best performing for this particular use case.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
This project aligns with the UN SDG 3, Good Health and Well Being. Medication prescription is an integral part of patient care in healthcare systems. It involves decision-making about the choice of medicines, its communication to pharmacists in the form of prescriptions for dispensing and finally, administration of medicines. The whole process requires seamless communication at various stages.
Incompleteness and illegibility of prescriptions account for a high proportion of medication errors that could potentially result in serious adverse effects. In a study by Jimma University Specialized Hospital, it was found that even though most of the physicians had knowledge on the importance of clear prescription writing, the magnitude of writing clear prescription was low because 54.8% of the sample prescriptions were proven illegible. More than 70% of the pharmacists in the study also reported that many physicians do not write legibly.
The study referred to above was in a system with professional handling each stage of the health system. In low resourced health systems where up to 90% of prescriptions are handled by nonprofessionals, illegibility of prescriptions is a bigger burden.
This solution will improve the better delivery of health outcomes by mitigating errors due to illegibility and improvement in digitization of handwritten prescriptions when the system is ready. Other solutions such as medication resolver that helps prescribers be aware of drug- drug interaction can also be built on its framework.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
The expected key outcomes are;
1. An AI OCR system that works well on hand-written medical prescription
2. An OCR system that works on hand-written medical prescriptions in low-resourced medical systems that accounts for those written in unstandardized formats such as plain sheets, torn sheets of papers
3. An easily adaptable tool that enables health workers to read and digitize hand-written prescription better.
4. Web and Android Application for online and offline OCR on hand-written medical prescriptions
5. An Application Programming Interface that other medical applications can connect with for OCR on hand-written medical prescriptions.
We have been able to solve the problem of training data availability by leveraging recurrent neural networks (RNN), a type of deep learning algorithm capable of text synthesis to generate human like prescription samples. These samples are available and can be presented.
There are also several views of different iterations of showing the process of AI enabled text extraction as the project is at this stage.
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
The design and thinking behind this project are inclusion and healthcare context specific implementation. Inclusion of handwritten prescriptions which are typically excluded when designing digital solutions related to healthcare. In the dataset synthesis, there was a coverage for not just global generic medication names but also national level medication brand names.
This work also explored the use of suitable synthetic data that is very similar to real prescription data and up to 12 different types of handwriting styles. Domain expertise of health workers was also considered in the project planning and execution.