AI-Based Early Lung Cancer Detection Using CT Scans With Multiple Classification
Project by Polygence alum Arif Barlas
Project's result
The project titled "AI-Based Multi-Class Lung Cancer Detection Using Deep Learning on CT Scans: A ResNet50 Approach for Early Diagnosis" yielded several significant outcomes: Development of an AI Model for Lung Cancer Classification: Utilizing the ResNet50 architecture, the project successfully created a deep learning model capable of distinguishing between various lung cancer subtypes—adenocarcinoma, large cell carcinoma, squamous cell carcinoma—and normal lung tissue in CT scans. High Classification Accuracy: The model achieved an impressive accuracy of 88% on the test dataset, indicating its potential effectiveness in clinical settings for early lung cancer detection. Comprehensive Performance Metrics: Detailed analysis of the model's performance revealed precision, recall, and F1-scores ranging between 90% and 95% across all classes, demonstrating balanced and reliable classification capabilities. Confusion Matrix Insights: The confusion matrix analysis highlighted the model's proficiency in correctly identifying most samples within each class, while also pinpointing specific misclassification patterns. This insight is crucial for guiding future refinements of the model. Training and Validation Consistency: Throughout the training process, the model exhibited consistent improvements in both accuracy and loss metrics, with minimal signs of overfitting. This consistency underscores the model's robustness and its potential applicability to diverse datasets. Ethical Compliance and Data Privacy: The project adhered to strict ethical standards by utilizing publicly available, anonymized datasets, ensuring the protection of patient privacy and confidentiality. Collectively, these outcomes suggest that the developed AI model holds promise as a supportive tool for radiologists, potentially enhancing the accuracy and efficiency of lung cancer diagnoses. However, further validation in clinical environments is necessary to confirm its robustness and reliability before widespread adoption.
They started it from zero. Are you ready to level up with us?
Summary
Cancer presents profound challenges due to the lack of effective treatment. Lung cancer, one of the most hazardous cancer types, causes 1.8 million deaths annually. However, if lung cancer in individuals who have died had been detected early, many of them might still be alive; therefore, early detection holds significant potential to save numerous lives. Moreover, emergent nations with limited healthcare infrastructure and insufficient availability of qualified radiologists, the likelihood of early detection is severely compromised. This study proposes the early detection of lung cancer using Artificial Intelligence (AI). Specifically, it employs multiple classification to distinguish between different subtypes of cancerous CT scans (adenocarcinoma, large cell carcinoma, squamous cell carcinoma) and normal scans. The ResNet50 architecture, pre-trained on the ImageNet dataset, is utilized for feature extraction and classification. ResNet50, pre-trained on the ImageNet dataset, was selected for its efficient feature extraction capabilities and ability to mitigate vanishing gradient issues, making it well-suited for processing complex medical images. The study achieved a noteworthy accuracy of 88 percent on the test set. The results demonstrate that ResNet50 can detect and classify cancer subtypes with a precision comparable to that of a radiologist, offering a promising tool to enhance diagnostic workflows, particularly in resource-limited settings. However, further validation in clinical environments is necessary to ensure robustness and reliability.
Krti
Polygence mentor
PhD Doctor of Philosophy
Subjects
Biology, Business, Computer Science
Expertise
Biology, Environmental Science, Climate sciences, marine ecology, environmental disease ecology, career mentorship, college mentorship, animal behavior, creative writing, financial success, Python, Java, R Studio, Google Earth Engine, Java, Start-ups for social good, and public speaking & leadership.
Check out their profile
Arif Barlas
Student
Graduation Year
2026
Project review
“Participating in this project has been an immensely rewarding experience, both professionally and personally. Developing an artificial intelligence-based model for the early detection of lung cancer provided a meaningful opportunity to contribute to existing healthcare services. By utilizing the ResNet50 architecture, we created a model capable of accurately classifying different cancer subtypes and normal tissues. This process underscored how deep learning techniques can revolutionize medical imaging. While the results are promising, indicating potential clinical applicability, we recognize the necessity for further validation and testing in real-world settings. This experience has highlighted the significance and excitement of working at the intersection of artificial intelligence and healthcare services. Additionally, the feedback on structuring my research and refining my presentation exceeded what I anticipated, making my findings more impactful and well-organized. The hands-on approach to troubleshooting technical issues and optimizing the AI model also met my expectations.”
About my mentor
“My mentor was extremely knowledgeable and supportive, offering clear and constructive feedback that helped me improve my research significantly. They were always available to answer questions and provided detailed explanations, ensuring I understood both the technical and practical aspects of AI in healthcare. Their ability to break down complex topics into digestible insights made the learning process much smoother. For any student working on AI applications in medical research, I highly recommend this mentor—they will challenge you to think deeper and push your project to the next level.”
Check out their profile