Advancements in Machine Learning and Data Science are not adequately reflected in how public health data is handled today. There is a visible gap between the advances in computing and medical sciences. In this position paper, we present an example of data science applied to the automation of a repetitive process within a cervical cancer screening program. We discuss the challenges for automating public health data and share our insights to elevate artificial intelligence (AI) in public healthcare.
More than 80% of the cervical cancer cases
and deaths in a year occur in low medium income
countries (LMICs) where prevention and cervical
screening resources are limited [1][
We aim to build a user-friendly automation that would allow medical experts to diagnose cancerous tissues of the cervix in a short period of time while reducing costs and technical experience required. This idea will work by combining heath and AI researchers’ expertise and experiences.
The main problem we aim to address is diagnosing biopsied women within a cervical cancer program. Our motivation is driven by the importance and time consumption of pathology process (i.e., pathologists reading histological slides). In the pathology process, women testing positive on screening tests are referred to specialised examination (colposcopy) to collect biopsy samples from the cervix and then haematoxylin and eosin (H&E) histological slides are prepared to be reviewed by pathologists using process (highlighted in yellow in Figure 1) excludes these steps for automation (highlighted in blue in Figure 1).
While the system works with the current process, the automation steps are currently done manually and repeatedly by a group of pathologists and statisticians. As the ML model does not exist in the current process, the analysis reports are produced after 2-3 stages of reviews involving multiple meetings to concur on the results. Including our proposed steps for automation in the current process will lower the burden of the experts and improve the timeframe up to 1/20 in comparison to the current process.
While our proposed project pipeline (Figure 1) forecasts optimal benefits for cervical cancer screening, in laying the groundwork, we were faced with critical challenges encompassing the realms of – technical, ethical, legal, and (most importantly) end user facing challenges. In this Workshop on “Healthy Interfaces (HEALTHI) 2022,” we look forward to discussing our research on automation of public health data analysis. We hope to share our current challenges, methods, and future plans for AI powered healthcare.
In this section we generalise the problems we faced when implementing our project (Figure 1) to discuss the challenges for automation of public health data analysis: 1. Trained Data Entry: The first challenge is the considerable effort needed to change the conventional data entry practices. To automate, it is essential to construct database constraints, design helpful interfaces, and train non-tech savvy workers to log: complete, error free, and rightly formatted data. 2. Patient Privacy: Anonymising the data is important to preserving the privacy of personal health records of patients who sign up for the study. If possible, it should be mindfully made visible at the level of the interface to both the patients and their clinicians. 3. Data Pre-processing: Data preprocessing is the cleaning and preparation of data for the model and analysis tasks. This is a time-consuming underestimated challenge, if done improperly, it potentially hinders the performance and accuracy of the model and delays the overall study. 4. Handling Large Datasets: As public health studies are ongoing processes which include participants on a rolling basis, they can result in large datasets during overall period of the study (which might span several years). It is crucial to prepare for handling the data in batches for faster training of the model. 5. Cross Validation by Experts: Validation of results is a necessity with respect to training ML models. In healthcare-related data, cross validation by experts is much more important to prevent fatal diagnosis errors and to check for any potential biases in the model. 6. Human Control: It is important to have adequate human control so that the confidence of the predicted results is higher. Enabling human control via the automation process’s interface allows to spot any discrepancies and malfunctioning. 7. Transparency: The interface should be made simple and transparent for both nonmedical and other non-tech savvy stakeholders involved. The entire automation process should be comprehensible to all stakeholders involved for the project to succeed. 8. Legal Efforts and Approval: Last but the most important challenge is to succeed in the legal efforts and approvals required for the automation projects. Developing proof of concepts with publicly available datasets is one of the ways to prepare for the challenge of gaining legal approvals and other grants
AI powered public healthcare will foster a health structure in the future where the AI process drives the speed and accuracy of the diagnosis, treatment, and recovery. People will get the right diagnosis at the right time such that their treatment and recovery chances improve, thus improving chances of a good life. Furthermore, the cost efficiency brought by AI techniques will enable smart healthcare to be adapted to different healthcare structures in different countries, specifically in the low-income countries, so that healthcare becomes accessible and affordable there. This is a possibility only when AI researchers combine their expertise and experiences with health researchers. With this position paper we aim to contribute by informing both medical professionals and computer scientists of the challenges for automation of public health data analysis.