Pancreatic cancer grading in pathological images using deep learning convolutional neural networks

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Pancreatic cancer grading is crucial for prognosis and treatment.
• Manual grading of pathology images is time-consuming and prone to misdiagnosis.
• Current methods lack efficiency and accuracy in determining cancer aggressiveness.

Purpose of the Study:

• To develop an automated system for grading pancreatic cancer using deep learning models.
• To compare the performance of different deep learning models on pathology images.
• To improve the accuracy and efficiency of pancreatic cancer grading.

Main Methods:

• A transfer-learning technique was employed using 14 ImageNet pre-trained models.
• Models were fine-tuned on a dataset of pancreatic cancer pathology images.
• Performance was evaluated based on classification accuracy for cancer grading.

Main Results:

• DenseNet models demonstrated superior performance in classifying pancreatic cancer grades.
• An accuracy of up to 95.61% was achieved in grading pancreatic cancer.
• The system showed high accuracy even with a small dataset.

Conclusions:

• This is the first study to grade pancreatic cancer directly from pathology images using deep learning.
• The proposed automated system can assist pathologists, improving diagnostic efficiency and accuracy.
• This approach overcomes limitations of manual grading and previous detection-focused or radiology-based studies.