Hyperspectral Imaging of Uterine Fibroids

Abstract

Uterine fibroids are non-cancerous growths of the uterus that affect nearly 70%–80% of women in their lifetimes. Fibroids can cause severe pain, bleeding, and infertility. The main risk of recurrence is smaller fibroids, which are notoriously hard to detect, being missed during a surgical removal procedure, only to enlarge afterwards. In this work, hyperspectral imaging (HSI) datasets were acquired from samples from 10 patients after receiving a hysterectomy. Optical properties including absorption, scattering, and spectral morphology were extracted and fed into machine learning to classify regions as fibroid and myometrium. Top extracted optical features had significant contrast between fibroid and myometrium (p < 0.0001) and were used to train Random Forest (AUC: 0.9985 ± 0.001, Sensitivity: 0.9534 ± 0.019, Specificity: 0.9936 ± 0.009) and Logistic Regression (AUC: 0.9397 ± 0.013, Sensitivity: 0.8405 ± 0.023, Specificity: 0.8895 ± 0.032) with strong performance across testing splits. With HSI, there is contrast between fibroid and myometrium in the human uterus.