EVALUATION OF LEARNING METHODS SIMILAR TO DEEP LEARNING AND DEVICE USING DEEP LEARNING FOR THE DIAGNOSIS OF THYROID NODULES

Abstract

INTRODUCTION
We recently developed a deep convolutional neural network algorithm (SEveRance Artificial intelligence program, SERA) using 13,560 ultrasound images of thyroid nodules labeled benign and malignant and this algorithm showed comparable diagnostic performance with experienced radiologists. We assessed whether the self-learning method similar to deep learning could be adapted for human learning as an ancillary approach to one-on-one education.

METHODOLOGY
Twenty-one internal medicine residents studied the “learning set” in three replicates which were composed of 3,000 images selected from 13,560 thyroid nodules and their diagnostic performances were evaluated before the study and after every learning session using the “test set”which was composed of 120 thyroid nodule images. The diagnostic performances of eight radiology residents were evaluated before and after one-on-one education using the same “test set”. After final test, all readers once again evaluated the “test set” with the assistance of SERA.

RESULTS
Before the study, the mean area under the receiver operating characteristic (AUROC) of internal medicine residents was considerably lower than that of radiology residents (0.578 and 0.701, respectively). Diagnostic performance of internal medicine residents, although not as much as radiology residents who received one-onone education (AUROC = 0.735), increased throughout the learning program (AUROC = 0.665, 0.689, and 0.709, respectively). All diagnostic performances of internal medicine and radiology residents were better with the assistance of SERA (AUROC 0.755 and 0.768, respectively).

CONCLUSION
A novel iterative learning method using selected ultrasound images from big data sets can help beginners learn to differentiate between benign and malignant thyroid nodules. With the assistance of SERA, the diagnostic performances of readers with various experiences in thyroid imaging could be further improved.