From Scalpel to Software: Artificial Intelligence in Autopsies
Written by: Taranjit Kaur, Medical Student & UCLA Post-Junior Pathology Fellow
Honoring the Foundations
Autopsies have been the gold standard in medicine for revealing the truth about the human body for centuries. Generations of doctors and medical students have learned about normal anatomy, rare disease patterns, causes of death, and the realities of effective and ineffective therapies through meticulous internal and external analyses of autopsy. Autopsies have also led to advancements in internal medicine and surgery, disclosed medical errors, and shaped public health policy. Burton and colleagues highlight autopsies as essential to medical training and quality development [1].
However, due to financial, logistical, and cultural challenges, autopsy rates have gradually decreased globally despite their invaluable use [1]. Artificial intelligence (AI) and other technologies are starting to transform postmortem practice as the field of pathology expands. AI might reinvigorate interest in autopsies and provide new diagnostic perspectives by addressing long-standing issues.
The Rise of Virtopsy
The term "virtopsy," which combines the words "virtual" and "autopsy," describes a minimally invasive or non-intrusive substitution for traditional postmortem practices. The virtopsy technique was first created in Switzerland by Professor Richard Dirnhofer and has become more prevalent in forensic contexts worldwide. Virtual autopsy, which does not require dissection, can provide full internal and external representations of the deceased body utilizing imaging modalities such as multi-slice computed tomography, magnetic resonance imaging, and 3D surface scanning [2-3]. By blending computer technology, forensic medicine, and radiography, virtopsy provides a contemporary alternative to classic autopsy methods.
According to Dass and colleagues, virtopsies have several advantages: They may be utilized safely in highly contagious or radioactive situations, require fewer dissections, and can be more welcoming due to their non-invasive nature. Furthermore, it allows experts to re-analyze cases years later by storing data, improves the consistency and speed of forensic exams, decreases staff exposure risk, and preserves human tissue [2]. However, some limitations to consider include the high cost, as not all institutions have access to AI, as well as variability in image resolution.
Educational Applications of AI in Pathology
Beyond virtopsies, AI may greatly enhance medical students' and physicians' education and training. Students could study independently, thoroughly examine normal and abnormal anatomy, and virtually practice dissections. They may also be able to save screenshots directly within the platform to review and revisit content later. Since incisions in real life are irreversible, it could be especially beneficial for developers to include undo and redo options, allowing students to practice their technique and decision-making in low-risk settings. Also, AI may have the capacity to safely store data online, which may facilitate cross-institutional collaboration to obtain second opinions.
Virtual Staining and Digital Histology
In addition to virtual autopsies, AI now can virtually stain tissue in controlled lab settings. Conventional histology staining methods are expensive, time-consuming, and especially susceptible to autolysis-induced deterioration in autopsy cases. To address these challenges, experts from UCLA Samueli School of Engineering created a novel deep-learning technique to process and stain autopsy specimens digitally. This unique deep-learning technique colors unstained tissue samples and generates high-quality images online without dyes or glass slides. Mechanistically, this all occurs by an engineered neural network, which is programmed to interpret autofluorescence images of tissue and subsequently generate images that replicate Hematoxylin and Eosin (H&E) staining. H&E is a key stain used on most tissues to highlight their architecture and morphology [4].
In a series of blinded evaluations and validation studies, pathologists from UCLA and partnering institutions analyzed tissue samples prepared using both standard and virtual H&E staining techniques without prior knowledge of which method had been used. They found that in well-preserved specimens, the two images appeared nearly identical. Interestingly, the virtual staining approach often revealed more distinct structural features than traditional staining in tissues affected by autolysis [4]. This demonstrates that virtual staining may be utilized over traditional staining when tissue preservation is poor.
Furthermore, while this innovative technology currently focuses on reproducing H&E staining, it leads toward a future in which pathologists may be able to digitally apply various special stains or immunohistochemistry (IHC) markers to tissue obtained from a biopsy or autopsy. When only a small amount of tissue is available in the lab, there are limited opportunities to order stains that provide diagnostic value. Once the tissue is used up, it cannot be restored. Thus, each decision a pathologist makes is critical. Future virtual staining could overcome limitations by allowing pathologists to apply a variety of stains to the same specimen without consuming tissue. In addition to increasing sustainability, reducing costs, and potentially lowering laboratory workload, this innovation also enhances diagnostic flexibility.
Ethical Considerations and the Future of AI
As AI progresses, it is critical to consider the implications beyond the laboratory. Careful oversight is paramount as AI technologies expand and integrate into autopsy procedures. Autopsy results can impact legal proceedings, epidemiological statistics, and vulnerable families seeking closure. Although AI can simplify workflows and emphasize findings, it should be considered a tool that supplements, not replaces, expert medical knowledge. Consistent with osteopathic principles of treating the whole person, AI should be integrated in a way that enhances clinical decision-making without compromising the physician’s role in delivering compassionate care [5].
AI can make autopsies more efficient, accessible, and reliable. Also, medical training can be enhanced. However, AI algorithms must be designed carefully, all findings must be thoroughly validated, and steps must be taken to avoid any diagnostic errors. For example, an AI algorithm may misinterpret a spill on a tissue slide and mistakenly diagnose it as an abnormality associated with poor patient prognosis. This emphasizes the significance of human oversight, continuous software updates, and constant surveillance of AI.
Ultimately, the primary responsibility for diagnosis remains in human hands. In the end, grieving families look to a pathologist, not a machine, for answers about their loved one's final moments.
References:
[1] Burton, J. L., & Underwood, J. (2007). Clinical, educational, and epidemiological value of autopsy. The Lancet, 369(9571), 1471–1480. https://doi.org/10.1016/S0140-6736(07)60376-6
[2] Dass, G., & Srivastava, A. K. (2021). Virtopsy: The new tool in the forensic science. International Journal of Advance Research and Innovative Ideas in Education, 7(1), 472- 474. https://ijariie.com/AdminUploadPdf/Virtopsy__New_tool_in_Forensic_Science_ijariie13527.pdf?srsltid=AfmBOopO9QyevjcUel88Rpn_S7PoRF3BdjahRciLvkIchgaf397kcSoa
[3] Badam, R. K., Sownetha, T., Babu, D. B. G., Waghray, S., Reddy, L., Garlapati, K., & Chavva, S. (2017). Virtopsy: Touch-free autopsy. Journal of Forensic Dental Sciences, 9(1), 42–46. https://doi.org/10.4103/jfo.jfds_7_16
[4] UCLA Samueli School of Engineering. (2024, March 6). UCLA unveils AI-powered autopsy tissue-staining method using deep learning. https://samueli.ucla.edu/ucla-unveils-ai-powered-autopsy-tissue-staining-method-using-deep-learning/
[5] Still, A. T. (1899). The philosophy and mechanical principles of osteopathy. Hudson-Kimberly Publishing Co.
