Multiscale Mask R-CNN-Based Lung Tumor Detection Using PET Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Positron emission tomography (PET) is crucial for diagnosing lung tumors but leads to radiologist overload.
• Computer-aided diagnosis (CAD) offers a solution to reduce workloads in medical imaging interpretation.

Purpose of the Study:

• To develop and evaluate a novel computer-aided diagnosis method for lung tumor detection using PET imaging.
• To leverage a multiscale Mask Region-Based Convolutional Neural Network (Mask R-CNN) for enhanced detection accuracy.

Main Methods:

• Three Mask R-CNN models were fine-tuned on PET images from different scales, each containing lung tumors.
• A weighted voting strategy integrated these models to minimize false-positive results.
• The method was tested on 134 PET slices.

Main Results:

• The proposed method achieved high performance metrics: precision of 0.90, recall of 1, and F-score of 0.95.
• Experimental results demonstrated significant effectiveness in detecting lung tumors.
• The system showed capability in identifying healthy chest patterns and reducing misidentifications.

Conclusions:

• The multiscale Mask R-CNN-based method is highly effective for lung tumor detection in PET imaging.
• This approach shows promise in assisting radiologists, improving diagnostic efficiency, and reducing errors.
• The study highlights the potential of AI in managing the increasing demand for PET scan interpretations.