Kidney Tumor Detection and Classification Based on Deep Learning Approaches: A New Dataset in CT Scans.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Kidney tumor (KT) is a prevalent global cancer, ranking seventh in incidence worldwide.
• Early detection of KTs significantly improves patient outcomes, reduces mortality, and enables effective preventive strategies.
• Traditional KT diagnosis is time-consuming and labor-intensive, highlighting the need for advanced diagnostic tools.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for the automatic detection and classification of kidney tumors in computed tomography (CT) scans.
• To assess the performance of novel 2D-CNN models against established architectures for KT diagnosis.
• To introduce a new, comprehensive dataset of kidney CT scans for research and development in KT detection.

Main Methods:

• Proposed three 2D-CNN models for KT detection: CNN-6, ResNet50, and VGG16.
• Developed a 2D-CNN model (CNN-4) specifically for KT classification.
• Utilized a novel dataset comprising 8,400 CT images from 120 patients with suspected kidney masses, split into 80% training and 20% testing sets.

Main Results:

• The CNN-6 model achieved 97% accuracy for KT detection.
• ResNet50 demonstrated 96% accuracy in KT detection.
• The CNN-4 model reached 92% accuracy for KT classification.

Conclusions:

• The developed DL models show high accuracy in detecting and classifying kidney tumors, offering a significant advancement over traditional methods.
• These models can reduce radiologist workload, minimize misdiagnosis risk, and enhance the overall quality of healthcare services.
• Early detection facilitated by these AI tools has the potential to alter disease trajectories and save patient lives.