OPTIMIZING KARYOTYPE ANALYSIS IN A REFERENCE CYTOGENETICS LABORATORY THROUGH WEB-BASED AI INTEGRATION

Karyotype analysis is a meticulous process that demands highly trained professionals and significant time for execution. Recently, artificial intelligence (AI)-based systems have been increasingly incorporated into cytogenetics to reduce analysis time and enhance productivity, offering a promising alternative for high-complexity laboratories facing rising test volumes. To validate and implement the VALI platform, a cloud-based system for image capture and cytogenetic analysis and to evaluate the platform's performance in processing and classifying bone marrow samples from patients with suspected acute myeloid leukemia (AML) and myelodysplastic syndromes (MDS). For validation, 50 onco-hematologic samples were processed. Following validation, from Sep/2024 to Apr/2025, a total of 3,046 onco-hematologic karyotype cases were analyzed using VALI. All cases were manually captured, and analyses were independently reviewed by two experienced cytogeneticists, ensuring diagnostic accuracy through mandatory dual validation. During validation, the average time for manual image capture was 42 minutes, and analysis time was 21 minutes. Among the 3,046 cases routinely processed, 72.9% showed normal karyotypes and 15.8% had relevant abnormalities and the remaining 11.3% were classified as ATM (total metaphase absence). A focused evaluation of 203 samples for suspected AML or MDS, received up to 48 hours after collection, showed that 140 (89.9%) were classified as IMA (>18 metaphases), 48 (23.6%) as BIM (3-17 metaphases), and only 15 (7.4%) ATM. Among the IMA and BIM samples, 55 (39%) and 15 (31%) presented relevant cytogenetic abnormalities, respectively (chi-square p = 0.320). The most frequent findings in karyotype were related to chromosomes 5, 7, 8 and 20. Overall, 93% of these AML/MDS samples were successfully classified using immunophenotyping and cytogenetics. The implementation of a web-based system with AI-powered chromosome clustering, remote access, and no licensing restrictions significantly increased analytical capacity, reduced turnaround times, and enabled decentralized karyotype analysis. The system supported remote access for image capture and interpretation, and maintained high technical standards through mandatory dual review by experienced professionals. Moreover, it allowed high-quality processing of AML and MDS samples, with a lower ATM rate than typically reported in the literature.