Postoperative Nocardia cyriacigeorgica infection after glioblastoma resection: correlating metagenomic next-generation sequencing with conventional microbiology.

Postoperative nocardial infection after cranial surgery is rare and difficult to diagnose because Nocardia spp. grow slowly in conventional culture. Metagenomic next-generation sequencing (mNGS) can shorten the interval to microbial detection, but read abundance is influenced by commensal DNA and by contamination during sampling, laboratory processing, and sequencing. mNGS findings therefore require correlation with conventional microbiology and with the clinical context. An 84-year-old woman developed purulent surgical wound discharge with epidural empyema approximately 2 weeks after resection of a World Health Organization grade 4 glioblastoma and postoperative corticosteroid exposure. Revision surgery comprised extensive debridement, removal of the infected bone flap and fixation screw, and evacuation of the empyemas. Direct Gram and modified acid-fast stains of the operative specimen showed branching, acid-fast filamentous organisms on day 1. mNGS of abscess fluid reported 196,785 reads assigned to Nocardia cyriacigeorgica on day 7, and culture on buffered charcoal yeast extract agar yielded N. cyriacigeorgica on day 18, confirmed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry. More than 20 million reads were assigned to Corynebacterium tuberculostearicum, yet no corresponding coryneform predominance was seen microscopically; its significance was therefore interpreted cautiously as a possible skin-derived or background contribution, although a true polymicrobial contribution could not be excluded. Surgical source control with intravenous meropenem and amikacin was followed by resolution of the infection. Additionally, three multimodal artificial intelligence systems were each queried five times with the same prompt. mNGS enabled earlier species-level recognition of N. cyriacigeorgica than culture, but direct microscopy gave the earliest diagnostic clue and culture remained essential for confirmation and isolate recovery. Taxonomic read abundance must not be equated with causality. The artificial intelligence outputs were discordant between systems and, in one system, incorrect in a stable and reproducible way; this illustrates current limitations rather than clinical readiness and supports the need for prespecified, blinded, multi-case validation before clinical deployment.