Blog, 26. 8. 2026

Brain tumors are one of the leading causes of death in the world [2]. Every year, thousands of lives are lost not because a cure didn’t exist, but because a diagnosis came too late. Brain tumors — whether glioma, meningioma, or pituitary — are among the most challenging conditions to detect, requiring expert eyes, precise imaging, and precious time. This project was born from a simple but powerful question: what if AI could help catch what human eyes might miss? By training a deep learning model to analyze MRI scans and classify tumor types, I set out to explore how artificial intelligence can move beyond the lab and into real lives — becoming a reliable partner in medicine’s most critical frontier, where every early diagnosis is a life given more time. 

What is a Brain Tumor?

A brain tumor is an abnormal growth of cells inside the brain. These tumors can be either benign (non-cancerous) or malignant (cancerous). Benign tumors usually grow slowly and may not spread, but they can still cause serious issues by putting pressure on surrounding brain tissues. Malignant tumors are more aggressive and can grow quickly, making them more dangerous.

This project specifically focuses on detecting three of the most common types of brain tumors:

• Glioma — A tumor that originates in the glial cells of the brain. It is one of the most common and aggressive types of brain tumors, often classified as malignant.
• Meningioma — A tumor that forms in the meninges, the protective layers surrounding the brain and spinal cord. It is usually benign but can still cause serious complications due to its location.
• Pituitary Tumor — A tumor that develops in the pituitary gland at the base of the brain. While mostly benign, it can disrupt hormone production and affect multiple body functions.

Brain tumors can affect normal brain functions and may lead to symptoms such as headaches, vision problems, difficulty in speaking, memory issues, and loss of balance. Because of these risks, early detection is extremely important for effective treatment and better patient outcomes.

Why Use AI for Detection?

Traditionally, doctors analyze MRI (Magnetic Resonance Imaging) scans to detect tumors. This process requires time, experience, and careful observation. AI can support this process by quickly analyzing images and identifying patterns that might indicate the presence of a tumor.

The goal of my project was to build a system that can assist in:

• Identifying whether a tumor is present
• Classifying the tumor type — glioma, meningioma, or pituitary
• Reducing the time needed for analysis
• Supporting medical professionals in decision-making

It is important to note that this system is not meant to replace doctors, but to assist them.

How the System Works

The system is based on a Convolutional Neural Network (CNN), a type of machine learning model designed for image analysis. Instead of manually programming rules, the model learns directly from data.

In simple terms, the process works as follows:

• The model is trained using MRI images labeled as glioma, meningioma, pituitary tumor, or no tumor
• It learns the distinct visual patterns associated with each tumor type
• It then predicts the result for new, unseen images

This learning-by-example approach makes CNNs very powerful for medical imaging tasks.

Project Implementation

To build this system, I followed a structured approach starting from data collection to model evaluation. The dataset consisted of MRI images labeled across four categories: glioma, meningioma, pituitary tumor, and no tumor. Before training the model, I performed data preprocessing to ensure consistency. This included resizing images, normalizing pixel values, and applying small transformations like flipping and rotation. These steps helped improve the model’s ability to generalize and perform well on new data. After preparing the data, I built the CNN model. The model extracts important features such as shapes and patterns from the images and uses them to classify each scan into one of the four categories. Once trained, the model was tested on unseen data to evaluate its performance.

Results and Observations

The model showed promising results with an accuracy of 96% and was able to correctly classify many MRI images across all three tumor types. One key observation was that the quality of the dataset had a major impact on performance. Better and more balanced data — particularly having sufficient examples of each tumor type — led to improved accuracy. Another important takeaway was that even a relatively simple model can produce meaningful results if trained properly. However, continuous improvement and testing are necessary to achieve reliable performance.

Challenges Faced

Working on this project came with several challenges. One of the main issues was the limited availability of high-quality, well-labeled medical datasets covering all three tumor types. Additionally, improving the model’s accuracy required multiple iterations and careful tuning. I also faced the challenge of overfitting, where the model performs well on training data but struggles with new data. Solving it required numerous adjustments in training strategy and validation techniques.

Conclusion

Brain tumors do not wait — and neither should our solutions. This project proved that AI has a real and meaningful role to play in medical diagnosis, not as a replacement for human expertise, but as a powerful extension of it. Detecting glioma, meningioma, and pituitary tumors from an MRI scan is just the beginning. The real mission is bigger, bolder, and deeply human: build technology that fights for lives. Every iteration, every challenge, every line of code — it all serves that one purpose. 

References:

[1] Mustafa, O., Alhatamleh, S., et al. (2026). A deep learning based NeuroFusionNet approach for automated brain tumor diagnosis from MRI. Frontiers in Neuroinformatics, 20. https://doi.org/10.3389/fninf.2026.1795354

[2] Bhimavarapu, U., Chintalapudi, N., & Battineni, G. (2024). Brain tumor detection and categorization with segmentation of improved unsupervised clustering approach and machine learning classifier. Bioengineering (Basel), 11(3), 266.https://doi.org/10.3390/bioengineering11030266

About the Author

Mahnoor Jamil is a researcher in Cybersecurity and Artificial Intelligence, holding 60+ certifications and recognition badges from leading organizations including Microsoft, Google, Cisco, and IBM. She is also an international speaker who has mentored 500+ women in cybersecurity and AI, actively contributing to the growth of an inclusive and skilled tech community.