Artificial intelligence Vs Radiologists

 Pneumonia is still one of the major causes of child mortality in Pakistan. It is not because of the lack of machinery and equipment in Pakistan, but it's because there is a lack of health officers and doctors to accurately and quickly read and transcribe those X-rays and advise the patients.

So I raised the question.

Can artificial intelligence, especially convolutional neural networks (CNNs), help radiologists detect pneumonia from chest X-rays more accurately and efficiently?

In a short time span his question becomes my independent research project, where I can research and think out the possible ways in which I can transform how we think about radiology in an underdeveloped country like Pakistan.

Why Pakistan specifically? 

You all will be inquisitive about it; that's why I specifically chose pneumonia for my research, which is because Pakistani children under 5 are more exposed to this disease in middle- or low-income families of Pakistan.

In this stance X-rays are the major diagnostic element. Where the radiologists are trained for this, which is lacking in Pakistan.

Even in the metropolitan cities of Pakistan where they are available, they are sometimes overloaded with work and sometimes overworked. In short, they lack time in each case.

One missing shadow and one overlooked opacity this is where it takes a delay, which continues to cost life.

What Is a CNN (Convolutional Neural Network)?

A CNN is a deep learning algorithm trained to “see” images like a human eye—but much more keenly and smartly.

It can:

• Scan more than thousands of chest X-rays in less time

• Learn to recognise patterns of pneumonia in preexisting X-rays

• and make predictions based on those patterns, which radiologists find hard.

In fact, CNNs have already been shown to match accurately or even outshine radiologist doctors in certain areas of this field.

What My Research Investigates

Primarily I am not interested in whether AI can detect pneumonia or not; what I want to know about this topic to fill the research gap is

1. How accurate is it really?

2. Does it speed up the diagnostic process?

3. Can it support and assist Pakistani radiologists?

To explore this, I’m analysing how CNNs (like CheXNet, a model trained on over 100,000 X-rays) compare to human radiologists when it comes to:

• Accuracy in terms of sensitivity and specificity
• Time taken to flag cases
• Confidence levels in decision-making.

Why This Matters in Pakistan

As of 2025, according to World Bank & PBS estimates, it’s estimated that more than 60% of Pakistan’s population lives in underdeveloped or rural areas, many of which lack access to basic healthcare, education, and infrastructure.

When there are no doctors at all, a chest X-ray might be taken and not read on time, and the worst-case scenario is that it will be misread by some undertrained staff.

The structure I developed that could come in working is that

• Local clinics could use AI as a first screener.

• Urban hospitals could use it as a second opinion tool.

• Governments could use it to track pneumonia trends across districts in real time. (Since the number of patients in government hospitals is higher)

Just delve into the scenario where an app where a nurse will upload an X-ray will tell her that this one looks abnormal and to send it to the hospital immediately.

My Initiative: ScanForTruth

Alongside my research, I’m also launching an educational page on Instagram called @ScanForTruth, where I break down:

• How AI reads X-rays

• Real case studies simplified

• What every student and patient should know about AI in healthcare

Because AI isn’t just for experts only, it’s for all of us.