A literature review of artificial intelligence (AI) for medical image segmentation: from AI and explainable AI to trustworthy AI

Abstract

Background and objective: Medical image segmentation is a vital aspect of medical image processing, allowing healthcare professionals to conduct precise and comprehensive lesion analyses. Traditional segmentation methods are often labor intensive and influenced by the subjectivity of individual physicians. The advent of artificial intelligence (AI) has transformed this field by reducing the workload of physicians, and improving the accuracy and efficiency of disease diagnosis. However, conventional AI techniques are not without challenges. Issues such as inexplicability, uncontrollable decision-making processes, and unpredictability can lead to confusion and uncertainty in clinical decision-making. This review explores the evolution of AI in medical image segmentation, focusing on the development and impact of explainable AI (XAI) and trustworthy AI (TAI).

Methods: This review synthesizes existing literature on traditional segmentation methods, AI-based approaches, and the transition from conventional AI to XAI and TAI. The review highlights the key principles and advancements in XAI that aim to address the shortcomings of conventional AI by enhancing transparency and interpretability. It further examines how TAI builds on XAI to improve the reliability, safety, and accountability of AI systems in medical image segmentation.

Key content and findings: XAI has emerged as a solution to the limitations of conventional AI by providing greater transparency and interpretability, allowing healthcare professionals to better understand and trust AI-driven decisions. However, XAI itself faces challenges, including those related to safety, robustness, and value alignment. TAI has been developed to overcome these challenges, offering a more reliable framework for AI applications in medical image segmentation. By integrating the principles of XAI with enhanced safety and dependability, TAI addresses the critical need for TAI systems in clinical settings.

Conclusions: TAI presents a promising future for medical image segmentation, combining the benefits of AI with improved reliability and safety. Thus, TAI is a more viable and dependable option for healthcare applications, and could ultimately lead to better clinical outcomes for patients, and advance the field of medical image processing.

Keywords: Medical image segmentation; artificial intelligence (AI); explainable AI (XAI); trustworthy AI (TAI).

Figure 1

Medical image segmentation based on AI. AI, artificial intelligence.

Figure 2

Some AI models: (A) an overall structure diagram of StillGAN; (B) an overview of TransUNet; the actual architecture is 3D. A 2D image is used here for the purpose of simplicity. AI, artificial intelligence; 3D, three-dimensional; 2D, two-dimensional; Conv, convolution; BN, batch normalization.

Figure 3

Some variations on existing structures: (A) Block of the Sharp U-Net; (B) testing pipeline of the SAM. W, width; M, mask; H, height; GT, ground truth; SAM, segmentation arbitrary model.

Figure 4

Medical image segmentation based on XAI. AI, artificial intelligence; XAI, explainable artificial intelligence.

Figure 5

Classification of XAI. XAI, explainable artificial intelligence.

Figure 6

LIME. (A) LIME interpretation of individual prediction process; (B) interpretations generated by LIME. LIME, local interpretable model-agnostic explanations.

Figure 7

Saliency maps. (A) Overall architecture of saliency maps. (B) Comparison of saliency maps represented as heat maps: (a) original image with polyp mask superimposed; (b) average expert; (c) average novice; (d) WM-DOVA energy map; (e) Bruce and Tsotsos model; (f) GBVS; (g) Itti-Koch model; (h) Seo model; (i) SIM; (j) SUN. High saliency areas correspond to hot regions in the image. WM-DOVA, Window Median Depth of Valleys; GBVS, Graph-Based Visual Saliency; SIM, Saliency Image Model; SUN, Saliency Understanding Network.

Figure 8

LRP. (A) Architecture of the model proposed by Aham et al. (B) The pipeline of the method, including the LRP module, the LRS module, and the classification module. LRP, layer-wise relevance propagation; LRS, lesion region segmentation.

Figure 9

IG. (A) Prediction of grade attribution of diabetic retinopathy from retinal fundus images. The original image is displayed on the left, and the properties (overlaid in grayscale on the original image) are displayed on the right. In the original images, the study annotated the lesions visible to humans and confirmed that the attributes pointed to them. (B) Comparison of IG and guided IG in the diagnosis of diabetic retinopathy. IG, integrated gradient; DR, diabetic retinopathy.

Figure 10

Counterfactual interpretation. (A) An overview of TraCE applied to the introspective analysis of CXR-based predictive models; (B) attribution framework. CXR, chest X-ray; H, height; W, width; D, depth; Conv., convolution.

Figure 11

Grad-CAM. (A) Architecture of Grad-CAM. (B) Architecture of C-CAM. Grad-CAM, gradient-weighted class activation mapping, C-CAM, causal class activation mapping; CNN, convolutional neural network; RNN, recurrent neural network; LSTM, long short-term memory; FC, fully connected; GS, global structure.

Figure 12

Benefits of XAI on medical image segmentation. XAI, explainable artificial intelligence; AI, artificial intelligence.

Figure 13

Medical image segmentation based on TAI. XAI, explainable artificial intelligence; TAI, trustworthy artificial intelligence.

Figure 14

Characteristics of TAI. TAI, trustworthy artificial intelligence.