With the increase in data in the healthcare industry, it doesn’t come as a surprise when Artificial Intelligence finds its way in the industry. In this article, we will take a deep dive into AI specifically its adoption in the healthcare industry. We will see the market trend and what the forecast for AI in healthcare looks like. We will get to know the existing penetration of AI and then we have a case study for Diabetic Retinopathy. In this particular case study, we will be using Computer Vision to solve a healthcare problem.
Computer Vision is the field in Computer science where we try to replicate the intricacies of the human vision. We will see how computers can gain high-level understanding from digital images or videos.
AI is transforming the Healthcare market, with its patients outcome which is aligned with the various stakeholders. It not only reduced the cost of healthcare but also provided a diagnosis solution to the people living in remote areas. For instance, Google DeepMind’s neural networks matched the accuracy of medical experts in diagnosing 50-sight-threatening eye disease including Diabatic Retinopathy, the case we are going to dissect and understand how exactly AI has been applied and their outcomes in this article. Various pharma companies are also using AI solutions for drug discovery. For example, Merck partnered with startup Atomwise and GlaxoSmithKline is partnering with Insilico Medicine.
Data from cbinsights
Healthcare AI start-ups’ received $4 billion in funding last year across 367 deals, according to intelligence platform CB Insights. That’s an almost 50% jump in cash from 2018, which saw almost $2.7 billion across 264 deals. Healthcare saw the most frenetic activity out of all U.S. industries, leading the finance and insurance sector ($2.2 billion across 198 deals) and retail and consumer-packaged goods ($1.5 billion across 159 deals).
Applications of AI in Healthcare
AI models have been very helpful when it comes to the Healthcare industry. There are multiple cases where AI has been able to diagnose the patient with greater accuracy and a lot faster, which helped patients with early detection. One such application is Cancer Cell detection using AI. Cancer is the most common risk that threatens human health worldwide. There are more than 100 types of cancer, including cancers of the breast, skin, lung, colon, prostate and ovaries. India had an estimated 1.16 million new cancer cases in 2018, according to a report by the World Health Organization (WHO), which said that one in 10 Indians will develop cancer during their lifetime and one in 15 will die of the disease.
Some other applications of AI in healthcare industry includes:
- Diagnosing deadly blood disease faster: Harvard University’s teaching hospital, Beth Israel Deaconess Medical Center, is using artificial intelligence to diagnose potentially deadly blood diseases at a very early stage.
- Brain Tumour Segmentation: Haveri et al., (2017) illustrated a brain tumour segmentation using deep neural networks to glioblastomas (both low and high grades) MRI image. This kind of brain tumour could appear anywhere in the brain with different shape, size and contrast.
- Cell Biology: In addition to medical science, artificial intelligence is also used for big data analyses in the field of molecular biology. Microscopic observation of cultured cells is important in cell biology. Specific cell types or conditions are recognized by fluorescently labeled antibodies. Each cell shows a characteristic gene expression pattern, including for structural proteins specific to the cell type and state; therefore, each cell type has unique morphological features.
More than 70 million people in India have diabetes and are at risk for diabetic retinopathy (DR), a vision-threatening disease. DR affects blood vessels in the light-sensitive tissue called the retina that lines the back of the eye. It is the most common cause of vision loss among people with diabetes and the leading cause of vision impairment and blindness among working-age adults. It doesn’t have any early symptoms. As of now, Retina photography is a way to detect the stage of Blindness. To spot the DR, there are 5 things medical expert spot on such retinal image-
Sign and Symptoms
Diabetics should seek treatment right away if they experience:
- Blurry, cloudy vision
- Floaters or dark spots in their field of vision
- Loss of central vision, especially when reading and driving
- Trouble driving or seeing at night
- Fluctuating vision
- Compromised color vision
- Vision loss
You are at high risk for developing diabetic retinopathy if you have type 1 or type 2 diabetes and:
- Have poorly controlled diabetes
- Have high blood pressure
- Have high cholesterol
- Are Hispanic or African American
Data Used for the Model
We are provided with a large set of high-resolution retina images taken under a variety of imaging conditions. A left and right field is provided for every subject. A clinician has rated the presence of diabetic retinopathy in each image on a scale of 0 to 4, according to the following scale:
0 – No DR
1 – Mild
2 – Moderate
3 – Severe
4 – Proliferative DR
Our task is to create a model that can predict these labels. Some sample images from our dataset are given below:
Challenges with the Dataset:
- The retina scan dataset has images across 5 different classes of diabetes level and the distribution of data is heavily imbalanced.
- Images come with many different lighting conditions, some images are very dark and difficult to visualize.
- Large uninformative areas.
Building a Computer Vision Solution
Advances in deep learning have improved AI. In medical, AI is mostly used as an alternative to human graders and particularly in the field of diabetic retinopathy it’s used for detecting retinal hemorrhage or classifying single photographs. Experienced ophthalmologists have gathered impressions of patients’ prognoses, but these impressions are not quantitative. Therefore, we trained a deep convolutional neural network using retinal images to autonomously perform the diagnosis. Below is the process flow illustration of the same:
We have used multiple architectures of Convolutional Neural Network like DenseNet, ResNet, Inception etc and created models which can predict the stages of Diabetic Retinopathy. We created models with different architectures and their training and testing scores are given below:
- In view of below 90% sensitivity and specificity of the device, 1 in 10 patients theoretically may have a false-positive and false negative result. A false negative result may provide a pseudo sense of security about the retinopathy status.
- It is crucial to educate patients and doctors that the present generation devices are not 100% reliable.
- Diabetes has numerous ocular manifestations other than DR, which includes glaucoma, age-related macular degeneration (ARMD), cataract, dry eye. A comprehensive examination is obligatory for proper diagnosis and management in these patients.
- Legal accountability in cases of misdiagnosis with artificial intelligence is another subject that is yet to be fixed.
In 2016, Google announced their inaugural work in training deep learning models for diabetic retinopathy (DR). Working with doctors at Aravind Eye Hospitals and Sankara Nethralaya in India, and also through their partnership with the Rajavithi Hospital, affiliated with the Department of Medical Services, Ministry of Public Health in Thailand, they are validating the model performance with patients from broad screening programs. You can read more about this study by following the source link.
Using Artificial Intelligence in the Healthcare industry can bring down the time and cost required for diagnosing the problem. At the same time, this diagnosis is not 100% accurate and if there is one false negative case that means one patient went without proper treatment which could’ve saved his life. At present, to use the benefit of AI in detection of Diabetic Retinopathy, we suggest the combination of AI for early detection and doctors consultation to remove any doubts. Surely AI is making progress by leap and bounds and we are moving towards the future where we can get diagnosis and treatment without any human intervention but we are not there yet and hence we should be diligent while relying upon these solutions.