Deep Learning Classification of Prostate Cancer on Confidently Labeled Micro-Ultrasound Images.
Journal:
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:
40039092
Abstract
Micro-ultrasound is a high-resolution ultrasound technology that has recently been introduced as an inexpensive alternative to MRI for prostate cancer identification. However, it is difficult to correlate micro-ultrasound imaging with MRI and ground truth final pathology due to tissue deformation from scanning and non-cartesian image orientations. In prior work our group has developed and validated a methodology for co-registration of micro-ultrasound and MRI with whole-mount pathology. Here we utilize this methodology to confidently label reconstructed micro-ultrasound images to train preliminary cancer classifiers with direct comparison against an expert micro-ultrasound reviewer. The trained models outperformed a novice reviewer and exhibited similar performance to an expert reviewer in a limited dataset of 15 patients (78.9% vs 60.6% sensitivity and 72.7% vs 80.5% specificity respectively). These results are encouraging and warrant further investigation with a larger dataset and more sophisticated models.