Deep learning models of ultrasonography effective in identifying superficial soft-tissue masses

1. In this retrospective study, the use of deep learning models improved the accuracy of radiologic diagnosis of benign and malignant superficial soft tissue masses. 

Evidence Rating Level: 2 (Good)

The majority of superficial soft tissue masses are benign tumors. Early detection and accurate diagnoses are necessary for determining treatment and prognosis for these masses. Diagnostic difficulties exist because there are over 70 different subtypes of benign superficial tumors, with few subtypes displaying the classic textbook signs. In this retrospective study, data were collected on 1615 patients with superficial soft tissue masses between January 2015 and December 2022. Two radiologists analyzed the ultrasound images of these masses and diagnosed them as either malignant or a subtype of a benign mass. Deep learning models (DLM) were also employed to analyze the images, with DLM-1 used to distinguish between benign and malignant masses, with an AUC of 0.992 (95% CI: 0.980, 1.0) and an ACC of 0.987 (95% CI: 0.968, 1.0), while DLM-2 was used to categorize the benign masses into the 5 most common subtypes. These subtypes are lipomyoma, hemangioma, neurinoma, epidermal cyst, and calcifying epithelioma, with AUCs of 0.986, 0.993, 0.944, 0.973, and 0.903, respectively. After referring to the DLM results, the radiologists re-evaluated their diagnoses. The initial diagnoses were then compared to the histopathological report for each mass. Using the DLMs, the radiologists improved their diagnostic accuracy. For the radiologist with 30 years of experience, using the DLMs improved their accuracy of diagnosing benign masses from 71.4% to 80.9% and malignant masses from 86.2% to 89.7%, while for the radiologist with 8 years of experience, the accuracy of diagnosing benign masses improved from 59.5% to 73.8% and malignant masses from 81% to 87.9%. A limitation of this study is that the radiologists only had access to images to make their diagnoses, while in clinical practice they can use information about patient history and presentation, which likely improves diagnostic accuracy. Overall, this study demonstrates that DLMs may be helpful in certain clinical settings for assisting radiologists with diagnosing soft tissue masses. 

Click to read the study in BMC Medicine

Image: PD

©2023 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.