Towards the Interpretability of Machine Learning Predictions for Medical Applications Targeting Personalised...

Area of Science:

• Medical Informatics
• Computational Biology
• Oncology

Background:

• Artificial Intelligence (AI), particularly machine learning (ML) and deep neural networks, is transforming medical applications.
• Cancer diagnosis and therapy are key areas benefiting from AI advancements.
• Current software tools require adaptation to meet the demands of AI integration in medicine.

Purpose of the Study:

• To survey current machine learning models and in-silico tools used in cancer research.
• To analyze the interpretability, performance, and input data of these AI models.
• To identify areas for improvement in AI for clinical practice.

Main Methods:

• Review of machine learning models including Artificial Neural Networks (ANN), Logistic Regression (LR), and Support Vector Machines (SVM).
• Analysis of Convolutional Neural Networks (CNNs) for image processing, leveraging Graphics Processing Units (GPUs) and High-Performance Computing (HPC).
• Evaluation of model interpretability, performance metrics, and data requirements.

Main Results:

• ANN, LR, and SVM are frequently used models in cancer research.
• CNNs are increasingly important for image-based cancer analysis.
• Model interpretability remains a significant challenge, hindering clinical trust and adoption.

Conclusions:

• Enhanced interpretability of AI models is essential for clinical decision-making.
• Improving AI interpretability will boost doctors' predictive capabilities.
• This advancement is key to achieving personalized cancer therapies in the future.