Artificial Intelligence and Cancer (Part 1): Brain Tumor Detection

The brain is your body’s central processing unit or CPU. Just like a computer’s CPU, the brain is extremely complicated in terms of its shape and function. The complications in this shape and function bring about difficulties in properly diagnosing brain cancer. However, researchers are looking into solving these difficulties through Artificial Intelligence. Scientists can utilize certain applications of artificial intelligence into image analysis. This can help identify if a specific tumor is forming in the brain. However, it is not as easy as it seems. Researchers still face challenges in hoping for practical use of artificial intelligence in our hospitals. 

The Need For Evaluating Brain Tumor

According to the Surveillance, Epidemiology, and End Results program, from 2015-2019, cancer in the brain and other nervous systems had a relative survival rate of 32.5%. This low survival rate is due to the lethality of later stages of brain cancer. However, when doctors can diagnose the disease earlier, the chance for treatment and survival increases. 

Similar to other types of tumors, brain tumors can either be malignant or benign. Along with the urgency of having a timely diagnosis, the right classification of malignancy of the tumor is also important. Still, some hindrances can also affect tumor analysis and classification. 

The degree of specificity of brain tumors is very high. The analysis must consider the size, shape, location, and brain tumor type. In addition, brain structures such as the cerebrospinal fluid, gray matter, white matter, and skull tissues all cloud an analysis of the tumor through an image.

Brain Tumor Imaging And Artificial Intelligence

Doctors commonly perform imaging of brain tumors through magnetic resonance imaging (MRI). A common problem among the images they produce is clarity. These images are prone to noise and artifacts that reduce image quality and clarity. This is where artificial intelligence (AI) comes in.

Contrary to what you might be thinking, using AI does not only mean analysis of the image to determine the type of brain cancer. AI can also be responsible for the pre-processing the image from an MRI and improving its quality. It does so by removing the noise and artifacts that impede accurate image analysis.

CNN: The AI for Image Analysis

After this image pre-processing follows the actual diagnosis of the tumor. With computer-aided detection, such as the use of AI, scientists can use several different algorithms and models to achieve an accurate reading. The most common model used is the convolutional neural network (CNN).

Researchers often use CNN because it is an AI network specializing in image processing. In collaboration with this algorithm, they may also use other models that utilize CNN’s image processing prowess. Some research attempts disregard CNN and instead specialize in reconstructing MRI images clearly. We will discuss all these chronologically to grasp a sense of development in the technology used for brain tumor detection. 

Before the Utilization of CNN

During the earlier years of AI integration in the medical field, the use of CNN was not that prevalent yet. Because of this, researchers focus on improving the image from an MRI for easier image analysis. From this, grew the interest in the image processing mechanism of AI. 

To demonstrate this image processing model, we turn to a 2018 study using mathematical morphological reconstruction (MMR) for image segmentation. This image segmentation boasts an improvement in superior image accuracy and a ten-fold decrease in computation time when compared to previous algorithms.

Although there was an improvement in image processing, the analysis of these images is still erroneous and inaccurate. However, recent developments in CNN could fix this problem.

Recent Advancements of CNN for Brain Tumor

CNN is the designated AI for image analysis. The algorithm is convolutional because it is an aggregation of several layers that each have their own function. These functions include input and output analysis, pooling, normalization, and many more, depending on the architecture used. 

In a 2019 paper that tested AI for brain tumor detection, scientists were able to aggregate 12 layers of CNN to analyze 226 test images of the disease. They used 1,666 images for the AI to train and learn brain tumor detection. This CNN method had an accuracy of 99.12%, a specificity of 96.42%, a sensitivity of 100%, and a precision of 98.83%. 

More and more researchers are modifying and optimizing their own CNN schemes. There was a study in 2021 that incorporated CNN into their own architecture called the correlation learning mechanism for brain tumor detection. This was only able to reach around 96% accuracy with 95% precision. However, the researchers added that the architecture learned faster. 

Another modification to CNN is the addition and subtraction of layers into the algorithm. A more recent study conducted in 2022 proposed a 13-layer CNN architecture. Compared to the 2019 paper, this study tested for 3064 images of three different types of brain cancer. Despite this increase in the number of images tested, it still achieved an accuracy of 97.2%. 

Regardless of these results and the ongoing efforts of our researchers to improve AI architectures, we still do not see applications of AI in the diagnosis of patients with brain tumors. The reasons for this lie in the complexity of brain tumors.

Challenges for Brain Tumor and Artificial Intelligence

The use of AI involves professionals in the field of engineering. This is one of the major flaws in the use of AI for medical services. As the manual tuning of parameters used in these algorithms is already time-consuming, it also requires extensive knowledge. Added to that, this process is also resource intensive in terms of the computers used. 

Ultimately, a doctor would still have to analyze the MRI image for a second opinion. Especially in the case of complex cancers,, this is exceedingly important. This would mean you would need to pay two professionals in different fields to diagnose a patient. The patient would have to shoulder this.

Practically, AI for brain tumors is still not feasible. However, when the time comes when engineers have created an interface for medical professionals to easily handle the AI, then it might become useful. When that time comes, researchers must have been able to sufficiently optimize AI in terms of time and resources.

Click here for our blog Disclaimer.