Early medical diagnosis and treatment of diseases is a domain where Artificial Intelligence (AI) has shown the potential to provide significant intervention.

Trained on vast troves of visual data of CT scans, X-Rays, digitized slides of cells, etcetera, AI based models have shown more accurate diagnosis than human pathologists in a very small fraction of time compared to diagnosis performed through traditional methods.

In United States alone, 1.9 million people are estimated to be diagnosed from cancer according to Centers of Disease Control and Prevention (CDC).

As the number of patients suffering from cancer is increasing every year, the number of pathologists in oncology to diagnose the cancer are present are very few, nearly a thousandth of the patients in need of diagnosis which leads to delay in. High quality of health care has led to reduce in number of deaths but those numbers can still be mitigated by practicing effective methods of early diagnosis.

AI has the potential to simplify the lives of doctors and patients by performing tasks that done by doctors in less time and at a fraction of cost.

Samples of 3 different classes of lymphoma cancer : 1. Chronic lymphocytic leukemia (CLL) 2. Follicelular lymphoma (FL) and 3. Mantle cell lymphoma (MCL).

The task of diagnosing the sub-type of lymphocytic cancer from the 3 classes CLL, FL and MCL by classifying the correct sub-type the patient is suffering from can be solved in short time by an AI model trained on large number of samples similar to input data with appropriate sub-type of each input image. The state-of-the-art accuracy for this task is 97.33 %.

Chest X-Ray images of people not suffering from pneumonia.

Chest X-Ray images of people suffering from pneumonia.

As it can be seen from the 2 figures above, the task now is to classify the X-Ray images of chest into two classes : 1. People suffering from pneumonia and 2. Those not suffering from pneumonia. The accuracy of an average radiologist is 85% while an AI based model achieves a classificatoin accuracy of 88.78%.

Such AI based models have not achieved 100 % accuracy yet which makes their deployment in real time scenarios risky. Instead, the results obtained from AI based models can be used as second opinion to the diagnosis made by the doctors. By practicing such methodology, doctors can cross check the diagnosis made by them with the results from AI models.

To facilitate this process there needs to be communication pathways between pathologists and clinicians for exchange of ideas.

This article provided a brief overview of early diagnosis using medical image analysis. From next articles onwards, I will dive deep into technical details of Artificial Intelligence, Machine Learning and Reinforcement Learning. References to all the information and facts has been provided below (papers I published during my senior year).

Thank you for reading.

This article was originally published by Sarang mahajan on medium.