Cytopathology is a minimally invasive, rapid method for early detection of cancer; however, inter-person subjectivity in reviewing cases can result in delayed or missed diagnoses. Existing methods that incorporate artificial intelligence (AI) to streamline the process do not yet meet clinical standards of accuracy and require human oversight. Nitta, Sugiyama, and colleagues developed a fully autonomous, clinical-grade cytopathology pipeline that generates high-resolution images coupled with AI-driven analysis. The edge computer–integrated optical whole-slide tomograph captures 2D images of samples across layers to acquire and compress high-resolution 3D images. This method consistently resolved subcellular structures across samples varying in thickness and from liquid-based cytology preparations. Real-time, AI-based population analysis of images by cluster of morphological differentiation (CMD) further enhanced diagnostic capabilities, capturing biologically meaningful transitions between morphological states. Cytological samples diagnosed as either negative for intraepithelial lesion or malignancy (NILM) or low-grade (LSIL) or high-grade squamous intraepithelial lesions (HSIL) were utilized to test the CMD framework, setting a threshold for positivity and generating plots to visualize spatial distribution of cells. To determine clinical validity, cervical cytology samples from 318 donors were processed and the platform’s ability to detect normal, LSIL, HSIL, and adenocarcinoma cells was assessed, in some cases outperforming human review by identifying potential underdiagnosis or false-positive results. Furthermore, the number of LSIL or HSIL positive cells were found to accurately stratify cases by severity. Utilizing this approach, samples were assigned as LSIL+ or HSIL+, and area under the receiver operating characteristic (ROC) curve analysis was conducted, with scores of 0.84 and 0.89 for LSIL+ and HSIL+ detection, respectively. Application of this workflow to cervical cytology samples from four additional medical centers demonstrated consistent performance, with ROC curve analysis values averaging 0.9 across sites. Overall, the whole-slide edge tomography and CMD-based analysis platform offers a promising automated clinical cytology pipeline with reduced subjectivity and increased scalability to capture high spatial resolution images of cytological slides, with powerful diagnostic value beyond current capabilities.Nitta N, Sugiyama Y, Sugimura T, Ito T, Ikebata K, Abe H, et al. Clinical-grade autonomous cytopathology through whole-slide edge tomography. Nature 2026;651:472–81.Note: Research Watch is written by Cancer Discovery editorial staff. Readers are encouraged to consult the original articles for full details. For more Research Watch, visit Cancer Discovery online at https://aacrjournals.org/cdnews.
A Fri, study studied this question.