How AI can impact medical diagnosis
The global medical diagnostics industry is growing at a steady rate and is expected to reach US$38.68 billion by 2025 at a CAGR (compound annual growth rate) of 50.2%. The Covid-19 crisis has highlighted the need to use technology to reduce the time needed to diagnose a disease and exposed the weakness of our healthcare systems.
The use of artificial intelligence (AI) in medical diagnosis can reduce detection time and error rate, use predictive techniques for self-diagnosis, better predict a patient’s future health and provide recommendations appropriate treatment.
Thanks to AI algorithms, we can predict the epidemiological trend and better diagnose the severity of the disease. Even though we have a huge influx of data from electronic health records (EHRs), patient histories, hospital records, and wearable devices, we don’t have enough technological capability to determine patterns and make predictions with precision.
To make reliable medical diagnoses, healthcare organizations leverage the power of high-quality data. Since data sources such as EHR, patient history, hospital records, and wearable device data are so diverse, maintaining data quality is a challenge. EHR data includes physiological data, lab test data, allergy information, patient insurance data, and medical history.
Data collected from patients and health records is rarely complete and clean and requires data cleaning. The cleaning step consists of removing errors, missing values and inconsistencies. Missing values can occur due to patient conditions unrelated to a particular healthcare variable or data not recorded by the sensor. If missing values are handled poorly, it can lead to biased results.
To handle missing values, we can either impute them or remove them. Before discarding missing values, we must ensure that the analysis of the remaining values does not produce inference bias. However, ignoring values is never the best option. Parameter estimation can be efficient, such as multiple imputations that impute missing values using Bayesian posterior distribution or K-Nearest Neighbor imputation.
Linear regression can calculate missing values when a variable distribution is continuous while logistic regression is useful for calculating missing values when the distribution is binary. Feature scaling helps to scale the features of data that varies in amplitude, range, and unit. Data scaling can be done using a min max or z-score algorithm.
As unstructured health data is generated rapidly from wearable devices, clinical reports, and medical prescriptions, prediction may not be useful due to irrelevant factors. Selecting the best features using feature engineering can improve accuracy and help identify the most important risk factors.
Principal component analysis (PCA) can help generate new uncorrelated predictors. For example, using PCA we can explore the main factors that increase the risk of heart disease. We can use the random forest algorithm to select the best feature subset.
The use of machine learning (ML) algorithms in medical diagnosis offers several advantages, such as early diagnosis of disease, reduction of treatment costs and the possibility of saving human lives. Initially, ML algorithms were designed to analyze huge sets of medical data. Over the years, we can now apply powerful machine learning algorithms, such as random forest and deep learning, to more specialized and complex medical diagnostic problems, such as heart disease prediction and classification. pictures of cancer.
A fuzzy logic system can be used for the diagnosis of heart disease. The system takes input variables such as heart rate, cholesterol, blood sugar, blood pressure (BP), gender, and age, and produces an output referring to disease state within a range numerical where increasing values indicate an increased risk of heart disease.
The fuzzifier changes the observation input to a fuzzy value. The fuzzy value is processed by an inference engine using a set of rules. The rule base in fuzzy systems contains a collection of attributes with AND/OR operators. For example, if BP, cholesterol and heart rate are low, the result is “Healthy”. Finally, defuzzification converts the output value of the inference engine into net logic.
Once we have developed and trained our model, we monitor the performance of our model to assess its performance on entirely new data. Certain performance criteria determine the success of the adoption of an algorithm in medical diagnosis. The most important measures used in clinical tests for the performance of medical diagnostics are sensitivity and specificity.
The predictive value of a test is defined as the likelihood of having the disease, given the test results. Positive predictive value or diagnostic accuracy is a measure of probability that indicates the number of cases that actually have the disease divided by the number of cases classified as disease by the classifier. A negative predictive value indicates the likelihood of a person being healthy when the classifier classified the output as negative/healthy.
The trade-off between high specificity (detecting all healthy people) and sensitivity (detecting people with disease) is important to consider in a binary class classification. We define threshold criteria to assign the input data to 0 or 1. Estimating the optimal threshold based on costs gives better results. For example, if the intervention of missing a disease is safe and cheap and the cost of diagnosis is high, the optimal threshold is at the top right of the ROC curve where the sensitivity is high and also the possibility of accepting a high number of false positives.
However, if the intervention is high risk and we are unconvinced of its effectiveness, the optimal threshold will be in the lower left corner of the ROC curve where we minimize harm to non-ill people but take missing sick people for granted. .
When we need to detect critical patients in healthcare facilities, it is important to prioritize sensitivity over specificity. In the case of someone positive for Covid-19, a false positive result may falsely suggest that a person is safe, and if they harm others, they may become a potential carrier of the disease.
Improving the personalized care of patients and the efficiency of care comes with ethical concerns. Since the reliability of the output is determined by the quality of the data entered, error-prone data can lead to ill-informed medical decision-making, which can ultimately impact health or cost the life of a patient. patient.
Another concern is that erroneous data may not correctly represent minority groups. This may put patients at risk of overdiagnosis or underdiagnosis. Characteristics such as age, disability, and skin color can also serve as the basis for algorithmic bias. For example, AI-based software that recommends skin cancer treatment to clinicians could be trained on white-skinned patients. Thus, this software will give inaccurate recommendations when testing samples containing information on, for example, African Americans.
Security and transparency are other crucial concerns in AI. AI developers should be transparent about the type of data used to train the model or any shortcomings in the model. For example, IBM’s Watson for Oncology provided incorrect cancer treatment recommendations because the developers had trained the software on synthetic cancer cases. IBM kept Watson’s incorrect and dangerous recommendations secret for more than a year. There can be a negative impact on the patient-provider relationship due to a lack of transparency. Healthcare providers must provide specific information about working with the third party for sharing patient data.
Recent advances in AI in healthcare have significantly contributed to medical diagnosis. With the rapid development of AI, although the focus is on optimizing the performance of complex AI models, there is a need for explainability. Explainable diagnosis will transform AI-based diagnostic potentials into clinical practice and provide a basis for reliable and reliable communication between AI model experts and medical experts.
AI-based systems will have a positive impact on several diagnostic fields, such as dermatology (interpretation of lesions), pathology (microscopic diagnoses), radiology (mammography or MRI evaluation) and ophthalmology (examination of an artery retinal for diagnosing diabetes).
Several startups are getting into the field of medical diagnostics using AI. Pharmaceuticals for the retina designed a device to diagnose glaucoma with accuracy and ease. Tricog, a Bengaluru-based startup, enables cardiologists to access blood flows and diagnose heart disease. OncoStem, another Indian startup, is developing new ways to detect breast cancer from a patient’s tumors and calculate the likelihood of recurrence.
Artelus, one of India’s top 10 artificial intelligence healthcare companies, helps doctors diagnose diseases by developing chest X-ray analysis products for the detection of tuberculosis (TB) and pneumonia. They have another product in their pipeline that will aid in the early detection of Parkinson’s disease and Alzheimer’s disease, which is currently in the data collection phase.
Can we deduce that AI will completely replace humans in the field of medical diagnostics? Our results suggest that this may not happen in the near future. AI has reduced the cost of prediction, but we still need audits and bias checks. We need experts to explain how a model arrives at a particular decision. We can say that with AI, we intertwine human knowledge with the intelligence of algorithms.
This article was published as part of Swasti 22, the Swarajya Science and Technology Initiative 2022. We invite submissions to the initiative.