Quantifying Post-Treatment Vascular Remodeling in Brain Aneurysms Using WEKA-Based Machine Learning: A Pilot Study

Ante Rotim^1,2Marina Raguz^3,4*Nikica Fulir¹Darko Orešković³Vladimir Kalousek⁵Petar Marcinkovic³Krešimir Rotim^6,7Bruno Splavski⁸Silva Soldo^2,9Tomislav Sajko^1,2

• ¹Department of Neurosurgery, Sestre Milosrdnice University Hospital Center, Zagreb, Croatia, Zagreb, Croatia
• ²Sveuciliste Josipa Jurja Strossmayera u Osijeku Medicinski Fakultet Osijek, Osijek, Croatia
• ³Department of Neurosurgery, Dubrava Clinical Hospital, Zagreb, Croatia
• ⁴Catholic University of Croatia, School of Medicine, Zagreb, Croatia, Zagreb, Croatia
• ⁵Department of Radiology, Sestre Milosrdnice University Hospital Center, Zagreb, Croatia, Zagreb, Croatia
• ⁶Special Hospital Neurospine, Zagreb, Croatia, Zagreb, Croatia
• ⁷Zdravstveno veleuciliste, Zagreb, Croatia
• ⁸Opca bolnica Dubrovnik Odjel za neurokirurgiju, Dubrovnik, Croatia
• ⁹Klinicki bolnicki centar Osijek, Osijek, Croatia

Abstract Introduction: To evaluate the feasibility of a WEKA-based machine learning pipeline for detecting post-treatment hemodynamic remodeling by comparing pre-and postoperative cerebral angiographic images in patients with middle cerebral artery aneurysms. Methods: This retrospective, single-center study analyzed 60 patients (51 women, 9 men; mean age, 58.2 ± 10.2 years) with unruptured middle cerebral artery aneurysms treated between January 2019 and June 2024. Thirty patients underwent microsurgical clipping, and 29 underwent endovascular intervention. A WEKA-based Random Forest classifier was trained on 15 manually annotated pre-and postoperative digital subtraction angiography (DSA) image pairs and then applied to the remaining dataset. Custom Python-based post-processing was used to denoise and refine the segmented images. Vascular surface area changes were assessed by comparing pixel counts before and after treatment. Statistical analysis included paired and unpaired t-tests, Mann-Whitney U tests, and effect size estimation. Results: Among 51 analyzable image pairs, 75% showed increased vascular pixel counts postoperatively, particularly in the endovascular group (segmented pixels: p = 0.034; refined pixels: p = 0.017). No statistically significant differences were observed in the neurosurgical group. Between-group comparisons of postoperative images did not reach significance. Conclusion: The WEKA pipeline enabled quantification of vascular remodeling but remained limited by manual preprocessing and lack of external validation. Machine learning–guided segmentation of angiographic images can detect treatment-induced vascular changes, particularly following endovascular therapy. This method demonstrates promise for future development of automated imaging biomarkers to support outcome monitoring and clinical decision-making in neurovascular care.