How to Detect human diseases using radiographs on machine learning models?

By Chirag
8 Min Read
How to Detect human diseases using radiographs on machine learning models? 1

The emergence of the latest technologies has completely transformed the world around us. Everything is more accessible than it was a decade back. Be it the field of business, education, e-commerce or anything else. Not only have processes become more simplified and hassle-free but also expedited. One such technology is machine learning that is transforming industries like no other for the good.

Machine learning has an immense impact on the world around us. Researchers are using it to find the answers to some of the most complex problems. The questions that took years to resolve, can now be answered in the least amount of time using some of the most profound machine learning algorithms. ML’s one of the best real-world impacts can be observed for the healthcare sector, where hospitals, organizations, clinics, and medical experts are working with software companies to develop comprehensive administrative systems and new drugs.

Not only this, but scientists are using machine learning to help diagnose and treat diseases in a much better manner. Some of these ML bases smart solutions are already being implemented in various parts of the world and helping patients recover at a rapid pace. In fact, ML models are also being used to improve the quality of life of people especially the elderly. Finally, the era of personalized medicine where the patient’s medical records and genetic information are studied to prescribe a course of action isn’t that distant anymore.

Why Radiographs?

Recent advances in medical science using machine learning predictions demonstrates that it is capable of more accurately detecting disease at an earlier stage, thus helping reduce the number of readmissions in hospitals and clinics. Radiographs are one of the most common medical images that are being used as an input for the ML algorithm. Since analyzing a radiograph is the primary criterion for diagnosing disease for a majority of diseases, researchers are capitalizing on it.

Another reason why researchers are using radiographs is that they are inexpensive as compared to other forms of diagnosis. Take a Python developing nation for an example, where a majority of the population find it difficult to spend money on availing healthcare services. They cannot get CT scans and MRIs along with other intrusive diagnosis methods performed since it requires a lot of money. But, when it comes to radiographs, they are not just inexpensive but also readily available at most of the physical diagnosis centers.

A radiograph records the images of internal structures of the body to access the presence or absence of diseases, foreign objects, along with structural damage or any anomalies. As a patient gets a radiograph, X-rays are passed through their bodies. While some of these passing X-rays are scattered by internal structures others are absorbed. The remaining X-ray patterns are then passed onto the detectors for later evaluation. Machine learning finds a great opportunity in taking its dataset as radiographs. 

Problem Formulation

The first and foremost step in detecting diseases or any abnormality is to formulate a problem and set a goal for what one craves. For example, radiographs can tell a lot about the internal structures, such as damaged bones, stable and unstable conditions of the structures along with others. But, when one is developing a machine learning model, there needs to be a specific set of rules that has to be followed. An ML model cannot haphazardly look for traces of multiple diseases at once but needs a precise goal it can reach.

Researchers can collaborate with hospitals and medical organizations to define a goal and build a model based on it. These goals can be one of the most sensitive and impending health issues for example. Alternatively, it can also be the one medical experts take time to diagnose and treat. While diagnosis is a crucial part of the treatment phase, any delay in it can cause the medical situation of a patient to escalate, even leading to death. As a result, ML models can be used to diagnose such problems that reduce the diagnosis time and aid the doctor quickly analyze the situation at hand.

Data Collection

The next part of the process is collecting the data or the radiographs in this case. With cloud platforms to the rescue, medical experts can collaborate with ML practitioners for receiving large datasets. Since hospitals and medical centers perform radiographs on patients they possess a large bank of data. This data can be then annotated by medical experts for relevant structures. For example, if doctors look at a vertebra to check if it’s fractured or not, medical experts must be able to annotate these vertebras for the model. Similarly, other crucial body structures that help a doctor make a decision regarding a patient’s well-being must be annotated. Researchers can take the help of medical annotation tools available in the market. Most of these usually create. Son files that can be shared as the input to the model.

Segmentation Techniques

Once the data collection part is over, the next thing to do is help the machine learning model identify the structures of a radiograph. A radiograph might capture a lot of data, but the ones that are annotated are considered relevant to detecting the disease. The segmentation phase of the project helps the ML model learn and then replicate the learning of the annotated structures. A part of the dataset is kept for validation while the others are used to train the model. The validation data is the one where the ML model replicates the learning that it has garnered from the training dataset.


The last part of detecting a particular disease or condition is building a classification model. Based on the number of inputs, researchers can select an appropriate model required for classification. The real challenge comes in deciding the number of deep layers that one must use for the task. Convolution neural networks are the best when it comes to analyzing medical images. Various classification models such as ResNet, AlexNET along with others can be replicated with the desirable input and an activation function. 


As researchers receive the results, these can be analyzed by the medical experts. The accuracy of the model can be increased by varying the number of neurons present in each input or varying the number of hidden layers in itself. Ultimately, the results will aid doctors and other medical experts make a faster decision in diagnosis. Put differently, it can also help other medical staff screen a patient based on the severity of the disease.

Share This Article
By Chirag
Software Programmer