An artificial intelligence-assisted diagnosis modeling software (AIMS) platform based on medical images and machine...

Area of Science:

• Artificial intelligence in medical imaging
• Machine learning for computer-assisted diagnosis
• Radiomics and deep learning applications

Background:

• Supervised machine learning, including radiomics and convolutional neural network (CNN)-based deep learning, is crucial for developing AI models in medical imaging for disease diagnosis and prognosis.
• Traditional machine learning workflows are complex, time-consuming, and require specialized expertise, posing challenges for radiologists and researchers in creating customized models for specific clinical needs.

Purpose of the Study:

• To develop a user-friendly software platform, Artificial Intelligence-Assisted Diagnosis Modeling Software (AIMS), that standardizes machine learning-based modeling workflows for medical image analysis.
• To provide an integrated solution combining both radiomics and CNN-based deep learning capabilities for comprehensive medical image analysis.
• To enable efficient model building, testing, and comparison through a modular design and graphical user interface, without requiring programming skills.

Main Methods:

• Developed AIMS, an all-in-one software platform integrating radiomics and CNN-based deep learning workflows.
• Implemented a modular design with a graphical user interface (GUI) for ease of use and accessibility for users without machine learning expertise.
• Included a flexible image processing toolkit for tasks such as semiautomatic segmentation, registration, and morphological operations to facilitate lesion labeling for various analyses.

Main Results:

• Demonstrated AIMS functionality in three radiation oncology experiments focusing on multiphase, multiregion, and multimodality analyses.
• Achieved an Area Under the Curve (AUC) of 0.776 for clear cell renal cell carcinoma (ccRCC) Fuhrman grading using multiphase analysis (n=187).
• Obtained an AUC of 0.848 for ccRCC Fuhrman grading with multiregion analysis (n=177) and an AUC of 0.980 for prostate cancer Gleason grading with multimodality analysis (n=206).

Conclusions:

• AIMS offers a user-friendly infrastructure that significantly lowers the barrier for radiologists and researchers to develop customized machine learning-based computer-assisted diagnosis models.
• The platform facilitates efficient and comprehensive medical image analysis, supporting diverse applications in disease diagnosis and prognosis.