Anatomy, as a foundational discipline in forensic medicine, has traditionally faced ethical concerns, limited access to cadaveric specimens, and inadequate simulation of pathological conditions (6). In recent years, advancements in computer vision, artificial intelligence, and medical imaging have enabled the development of novel virtual anatomy techniques. These techniques primarily involve the integration of multimodal imaging data to construct high-precision 3D anatomical models, which are further enhanced through interactive segmentation and physiological simulation capabilities (Figure 2). Building on these advancements in virtual anatomy techniques, the process begins by selecting organs relevant to forensic medicine teaching. Once selected, CT scanning is performed, generating a series of continuous, high-resolution two-dimensional cross-sectional grayscale images (7). The original image data exported from CT is converted into a universal format that MATLAB can read and process. MATLAB then stacks all the two-dimensional images in sequence and builds a 3D volume data matrix in memory. Subsequently, a continuous virtual 3D anatomical diagram composed of numerous images is directly generated through surface rendering. The image boundaries are then modified through image boundary fitting. Finally, a complete, closed, and grid-based 3D model of the anatomical organ is constructed by integrating 3D reconstruction technology. In the realm of 3D reconstruction, deep learning-based voxel modeling has emerged as the predominant approach for creating high-fidelity anatomical models. For instance, the modern VGGT model utilizes a pure feedforward transformer architecture to perform 3D geometric reasoning based on single or numerous input images (8). This research has been proven capable of quickly generating 3D imaging of the crowd at tourist attractions, which can be used for real-time monitoring of the population density at tourist sites. It also has great potential for use in forensic medicine teaching. However, no studies have yet utilized the VGGT model in forensic medicine teaching, and further research is needed in this regard. Another widely adopted method is multi-view stereo vision, which simulates visual disparites using two or more cameras and reconstructs image surfaces through feature matching algorithms (9). This algorithm not only can reconstruct a highly complete model, but also has low computational cost and good generalization ability. It can be applied to small-scale platforms with limited computing resources and is more suitable for forensic medicine teaching. However, the test data in the article is limited, so this result still needs further verification.

This potential for broader application is already being realized in contemporary practice. Virtual dissection and 3D reconstruction technologies are now being implemented in forensic practice and university education worldwide. Pringle et al. (10) explored the integration of 3D virtual imaging technology into forensic medicine education by establishing virtual anatomy laboratories, simulated outdoor environments, and interactive educational video games. The implementation of these technologies has contributed to the development of a sustainable and practical learning environment. However, the study was limited to a single class comprising 40 students, representing a relatively small sample size; thus, the findings require validation through larger-scale experimental studies. Similarly, Darras et al. (11) introduced virtual anatomy technology in anatomical instruction. Their results indicated that 78.7% of students perceived an enhanced understanding of cadaveric anatomy and its clinical applications through virtual learning. Although this study involved a larger sample of 202 participants, data collection relied solely on student self-reported surveys, without objective assessment of knowledge acquisition or skill development. Although 3D reconstruction technology is gradually being applied across many fields of forensic medicine teaching (Table 1), several significant limitations remain. First, a major issue lies in the deplorable adaptability of current systems to complicated pathological conditions. Given that forensic medicine involves a wide range of complicated cases where injuries vary significantly among victims, the virtual anatomical or pathological models generated by these technologies often fail to accurately represent complex pathological structures. Second, another limitation lies in that it is expensive to acquire high-quality medical imaging data and it is costly in terms of substantial computational resources to train 3D models. These high costs hinder their extensive adoption, particularly in resource-limited universities (12, 13). Finally, because virtual dissection lacks the tactile feedback provided by physical dissection, students may develop an incomplete appreciation for the sanctity of human life (14).

Traditional teaching process in forensic medicine education largely relies on case-based instruction. However, due to constraints such as case confidentiality, fragmented resource distribution, and limited hands-on practice, forensic medicine teaching often struggles to access real case materials, involves high operational risks, and suffers from low-fidelity scene reproduction (25, 26). With advancements in artificial intelligence and virtual reality, a dual-core teaching system combining “intelligent case repository” and “simulation teaching system” has emerged (27, 28). In the future, forensic medicine teaching can also utilize this technology to collect real case data and establish three-dimensional models. Then, these models are used to develop a comprehensive platform for forensic medicine teaching that encompasses forensic pathology, forensic evidence examination, and crime scene reconstruction.

At the core of this system is the construction of a library of forensic intelligent cases that relies on the collection and standardization of multimodal data. Information from documented past cases—such as case descriptions, autopsy images, scanned physical evidence, and on-site investigation records—should be integrated into the database and annotated using multi-label classification techniques (29). Concurrently, an interdisciplinary knowledge graph should be developed to connect knowledge from anatomy, pathology, and forensic medicine, that can enable cross-case knowledge transfer and supporting cross-case reasoning. The theory of experiential learning emphasizes learning through a cycle of specific experiences, reflective observation, abstract conceptualization, and active experimentation. However, fully implementing this cycle in traditional forensic education has been challenging due to the constraints of reality, resources, and safety. AI technology, emerging in this context, provides an unprecedented path to realize this theoretical model. The system that simulates virtual reality crime scene can generate an infinite variety of virtual scenarios. Under this system, students can immerse themselves in a highly realistic 3D environment by wearing VR equipment and conduct on-site investigation practices.

In teaching, the intelligent case library should be integrated with a simulation-based teaching system. Such a system comprises three main components, including the data layer, the algorithm layer, and the application layer (Figure 3). The database of the future forensic medicine simulation teaching system can be built upon the foundation of the forensic intelligent case database, and it includes a three-dimensional model library and a teaching resource library. The algorithm layer serves as the core technical component of the simulation-based teaching system, primarily utilizing a combination of “virtual reality engine + AI reasoning engine + behavior analysis module” to enable dynamic interpretation of case scenarios in a virtual environment. The algorithm layer can provide simulated crime scenes for forensic education and analyze the students’ operational processes during the on-site investigation. Finally, the application layer comprises tools such as a virtual autopsy table, an evidence analysis simulator, and a scene reconstruction sandbox. These tools bridge real case data with immersive simulation, allowing trainees to perform virtual autopsies using haptic feedback devices. Trainees’ procedural steps and diagnostic reasoning are then synchronized and compared with AI-generated reference solutions. This enables the automated generation of multidimensional evaluation reports to assess student learning outcomes (30, 31). Currently, several institutions have begun developing forensic simulation teaching systems. Mayne et al. (17) developed a virtual crime scene investigation laboratory using 3D imaging technology, enabling students to wear head-mounted display devices and conduct immersive investigations within a simulated crime scene. Participants reported high levels of satisfaction, and notable improvements in forensic investigation skills were observed. Nevertheless, the study included only 10 cases, limiting statistical robustness, and some students experienced adverse physical reactions such as dizziness during the experience. Darlene et al. (24) implemented virtual imaging technology to create a virtual autopsy laboratory, where students demonstrated improved proficiency in autopsy procedures, with nearly 80% achieving competent performance. This study shared similar limitations with prior research, including a small sample size and reports of motion-related discomfort among users. Collectively, these studies indicate that simulation-based teaching systems are increasingly being adopted in forensic medicine education and show promise in enhancing learning outcomes. However, due to current technological constraints, virtual reality remains insufficient to replace traditional hands-on training and should be regarded as a supplementary tool rather than a primary instructional method. Widespread adoption as a core teaching approach is still distant.

The constructivist learning theory holds that knowledge is not passively received but is actively constructed by learners based on their experiences (34). The introduction of AI also promotes this active process of knowledge construction. With the continuous AI development, the educational model in forensic medicine in the future will evolve from the traditional “teacher-student” binary structure to a “teacher-machine-student” tripartite model. This shift signifies a transition from the knowledge-based “passive reception” paradigm to a personalized “interactive practice” one that integrates teaching, learning, and practical application, powered by AI. In the future, this interactive teaching model will have the following characteristics. First, AI can construct highly realistic virtual environments, overcoming time and space constraints. The integration of virtual simulation experiments into the curriculum enables students to safely and repeatedly practice skills like fingerprint collection and ballistic trajectory determination, thereby significantly enhancing their on-site investigation capabilities (35). Second, AI personalizes learning pathways through student data analysis and dynamic feedback, enabling “data-driven” personalized instruction. During the student learning process, AI generates customized training plans based on students’ operational data. Additionally, AI can generate virtual cases of varying difficulty levels, ranging from basic poisoning identification to complex multi-cause death analysis, catering to diverse and personalized learning needs. The theory of mastery learning holds that as long as sufficient time and appropriate conditions are provided, the majority of students can master the learning content (36). The AI-driven personalized learning system provides the operational framework to realize this principle in forensic medicine education. Third, AI allows forensic medicine to move toward practical, interdisciplinary collaboration with law, computer science, and data science by facilitating real-scenario integration. Similarly, authentic legal cases can be introduced as teaching examples, with AI providing structured guidance and performance evaluation for students.

The focus of forensic medicine courses has been on core domains such as forensic pathology, toxicology analysis, and evidence technology, while limited integration with fields like computer science and data science remained limited. The onset of AI has fostered interdisciplinary interaction between forensic medicine and fields like computer science and data science. This development necessitates forensic professionals not only to master forensic medicine knowledge but also to acquire expertise in computer science, biology, and law. To cultivate interdisciplinary professionals skilled in “medicine + technology + law,” modern forensic medicine teaching should incorporate foundational knowledge such as algorithm principles and data ethics. The iterative updating of AI-driven teaching content can be achieved through a dynamic closed-loop process of “technology evolution monitoring → case library update → teaching module adjustment” (37). The proposed curriculum reconfiguration will rest on a foundational framework comprising forensic medicine principles, biostatistics, and Python programming in the future. It will integrate core technical components such as forensic image processing, biometric recognition algorithms, and judicial big data analysis. The curriculum bridges classroom instruction and real-case scenarios through key integrative components such as virtual crime scenes and AI-assisted report writing. Furthermore, it is recommended to implement a “dual mentor system” involving both forensic medicine faculty and AI engineers in the teaching faculty.

At present, forensic medicine education assessment is primarily built on final written exams and experimental reports (38). However, this approach struggles to evaluate critical forensic competencies such as students’ capacity for critical reasoning in crime scene reconstruction and the hypothesis verification process in forensic evidence analysis. To address these limitations, it is urgently necessary to establish an intelligent assessment system based on multimodal data fusion. This system can adopt a paradigm of process-oriented assessment, namely, a shift from the traditional terminal exams to continuous competency profiling (Figure 4). This assessment requires the integration of biometric recognition, learning behavior modeling and knowledge graph tracking technologies (39). It enables a more comprehensive judgment of whether students have acquired the required forensic competencies. Its technical framework can be broadly categorized into three components, including a multimodal data collection layer, a dynamic assessment model, and the biometric recognition technology. The multimodal data collection layer can be constructed by integrating data from four sources: eye-tracking, speech semantics, experimental operations, and digital footprints. During the assessment, Biometric recognition technology is used to input the collected data into the dynamic assessment model. Then, a 2-D assessment framework based on a time-axis row and a space-axis column is established to serve as the foundation for subsequent comprehensive evaluation. Along the time-axis, the system records students’ learning trajectory, while on the space-axis, it maps the knowledge nodes they have mastered. By analyzing students’ operation trajectories, critical reasoning in decision-making, and time spent on different tasks within the virtual environment, the system provides a holistic evaluation of their progression from knowledge acquisition to skill application. Currently, this intelligent assessment system is still in the theoretical stage. Some technologies are not yet capable of meeting the requirements of this intelligent assessment system. However, in the future, it is certain to become a hot topic in forensic education research.

As AI seeks deeper integration into forensic medicine education, data barriers have emerged as the primary obstacles to leveraging AI in forensic medicine education. Due to the sophisticated and diverse nature of forensic medicine scenarios, it is difficult to collect certain forensic data such as corpse images, DNA samples, and crime scene records. Moreover, data annotation can only be performed by professional forensic experts, which increases both cost and subjectivity of the annotation process. The inherently poor interpretability (“black box” nature) of AI algorithms and the relatively low level of confidence and trust in AI also undermine the adoption of AI in forensic medicine teaching (40, 41). This field demands that conclusions be scientifically valid and legally defensible. However, current AI models often function as “black boxes,” obscuring the internal reasoning process and limiting both transparency and interpretability. This This opacity may lead to misjudgements caused by biased training data. Such misjudgements risk distorting students’ foundational understanding of forensic reasoning. Aydogan et al. (42) found that demonstrated that ChatGPT-4 produced forensic reports with high accuracy, albeit slightly lower than that of human experts, underscoring ChatGPT’s limitations in handling forensic cases. However, the cases used in the article were relatively simple. In reality, when dealing with complex forensic incidents, the accuracy of AI still needs further research. Finally, the limited infrastructure of many educational institutions, particularly in terms of hardware resources and computing power, further constrains the application of AI in forensic medicine education. Within these constrained applications, AI is primarily used to process and reconstruct high-resolution images in forensic medicine teaching environment. While current technologies can meet basic needs for image recognition, challenges remain in areas such as augmented reality systems, particularly in capturing high-quality images under poor lighting conditions.

AI systems in forensic education are increasingly attracting attention with regard to liability. Article 2 of China’s “Interim Measures for the Administration of Generative Artificial Intelligence Services” states that “industry organizations, enterprises, educational and research institutions, public cultural institutions, and relevant professional bodies that develop and apply generative artificial intelligence technologies but do not provide such services to the domestic public are not subject to these measures” (43). Compared to educational AI, operational forensic AI requires even higher standards. The EU’s “Artificial Intelligence Act” explicitly classifies AI systems used in judicial and law enforcement contexts as high-risk, mandating strict compliance with requirements such as risk management systems, high-quality datasets, activity logging, comprehensive documentation, transparent user information, human oversight, and robustness, accuracy, and cyber security standards (44). While educational AI is included within the broader regulatory framework, its specific obligations remain significantly more lenient, focusing primarily on data privacy and educational equity.

Data governance and privacy constitute fundamental challenges in the development of forensic education AI. Teaching in this domain depends heavily on real-world case materials, which often involve sensitive information related to the dignity of the deceased, personal privacy, and judicial confidentiality principles. To address these concerns, a tiered and categorized data governance framework must be established. First, clear legal norms should govern the custody and use of forensic data. Second, technical solutions such as data de-identification and synthetic data generation should be advanced to build simulation databases that protect privacy while preserving educational value. Additionally, formal consent mechanisms and flexible frameworks for the ethical use of human remains and case records in education must be developed to ensure both legal compliance and procedural transparency.

Algorithmic bias and limited explainability represent core risks in forensic education AI. Bias in algorithms may result in inaccurate assessments during training, while poor model interpretability undermines confidence in AI-assisted analytical conclusions. A dual strategy is required to mitigate these risks. First, institutions should strengthen the detection of algorithmic bias and assess its impact on educational outcomes by integrating algorithm auditing and fairness evaluation into curricula, enabling learners to understand how biases emerge. Second, efforts must focus on enhancing the explainability of AI models within judicial contexts. Through targeted instruction, students should develop the ability to critically evaluate the reasoning behind AI-generated outputs, assessing their reproducibility and interpretability according to evidentiary standards.

Teaching dependence presents a practical challenge in forensic education. Forensic medicine relies extensively on authentic case data for hands-on instruction. This reliance, coupled with restricted access to real data due to privacy and legal constraints, not only limits pedagogical effectiveness but also prevents students from experiencing realistic scenarios involving AI-assisted forensic practice. Only through access to sufficient, compliant simulated case data can AI technologies be effectively integrated into forensic education, thereby avoiding theoretical detachment from actual practice.

The erosion of professional judgment poses a potential crisis in the integration of AI into forensic education. Overreliance on opaque AI outputs during model development and deployment may lead to two critical issues: first, undetected errors may become entrenched and perpetuated, compromising analytical accuracy; second, practitioners’ capacity for independent professional judgment may deteriorate, fostering uncritical dependence on technology. To prevent this, forensic AI education should promote interdisciplinary curriculum design and ethical standardization, jointly establishing ethical guidelines and operational protocols for AI use. Furthermore, through case-based teaching and structured ethical discussions, students’ abilities in holistic judgment and professional accountability should be strengthened, ensuring that technological tools support—rather than supplant—professional expertise.

Since integration of AI into forensic medicine education necessitates collaboration across various disciplines, including medicine, computer science, and law, educators involved in forensic medicine must possess knowledge in these areas. However, this is compounded by a mutual knowledge gap: most forensic educators lack AI expertise, and computer scientists seldom grasp the nuances of forensic medicine. Johnson et al. (45) conducted a questionnaire survey among 161 staff members and graduate students engaged in forensic dentistry. The feedback indicated that dental professionals and graduate students had limited knowledge of AI technologies such as VR and 3D. This disconnect impedes the alignment of technology with pedagogical needs.

While exploratory projects in AI-assisted forensic medicine education have emerged at some institutions, the broader field still urgently requires the systematic development and expansion of interdisciplinary curricula across universities to cultivate professionals with the necessary interdisciplinary expertise.