Unsupervised machine learning identifies neurophysiological and anatomical correlates of the visual field deficits in patients with pituitary tumors
Pituitary tumors present oftentimes with a combination of visual field defects and diminishing visual acuity. Despite the frequency of these symptoms, the neurophysiological and anatomical factors that best predict visual field defects remain unclear. We applied unsupervised machine learning techniques to electrophysiological and imaging features in order to determine those features that determine the observed visual field defects.
29 Patients (median age=67, 10 female and 19 male) with non-functioning pituitary tumors where characterized so far retrospectively. For all patients that had a high-resolution MR of the sella, a formal ophthalmological evaluation in terms of visual field deficits (reported as mean deficit (MD) in dB) as well as measurement of visual evoked potentials (VEP) using LED pads we selected as variables of interest the latency and amplitude of the VEP, as wells as the chiasm angle, maximal and minimal height of the chiasm as evaluated independently by 2 observers on T2 MR Scans.
We observed 78% of visual field deficits defined as pathological (>3dB mean deficit) and in 60% of cases a visual field defect of at least 5dB. The mean chiasm angle (as measured in the coronar plane by drawing a line along the inferior edge from one side of the chiasm to midline and another from the midline to the contralateral side of the chiasm) was 121.87 degrees. The mean chiasma height was 1.9mm. Next, we performed a dimensionality reduction using principal components analysis (PCA) on the pooled data followed by k-means clustering along the dimensions with the 2 largest eigenvalue. The factors with the larges variance explaining the data where VEP latency (defined as the time from stimulus to N2 peak, p=0.02), the angle of the chiasm (p=0.00015) and the maximal chiasma height (p=0.02).
Dimensionality reduction using PCA coupled with a clustering algorithm (k-means) revealed that severity of the visual field deficits in patients with pituitary tumors is largely represented in neurophysiological features (VEP latency) as well as anatomical characters (chiasm angle). Out study suggest that intraoperative monitoring of these features may improve outcomes for patients undergoing surgery.
We use cookies on our website. Cookies are small (text) files that are created and stored on your device (e.g., smartphone, notebook, tablet, PC). Some of these cookies are technically necessary to operate the website, other cookies are used to extend the functionality of the website or for marketing purposes. Apart from the technically necessary cookies, you are free to allow or not allow cookies when visiting our website.