AI for Queensland Businesses Longtail Solutions

Introduction

In the world of healthcare, diseases like endometriosis can have severe consequences for patients. One such consequence is the obliteration of the Pouch of Douglas (POD) in the pelvis, which is often seen in patients with endometriosis. Diagnosing this condition accurately is crucial for effective treatment.

The Sliding Sign Test

The sliding sign is a dynamic transvaginal ultrasound (TVS) test that can help diagnose POD obliteration. This test involves the use of ultrasound to determine the state of the POD. However, interpreting the results of this test can be challenging and requires the expertise of trained sonologists.

Developing a Deep Learning Model

To overcome the challenges associated with interpreting the sliding sign test, we aimed to develop a deep learning (DL) model that could automatically classify the state of the POD using recorded videos of the test. This DL model would be able to analyze the videos and accurately determine whether the sliding sign was positive (normal) or negative (indicating POD obliteration).

Methods

To develop the DL model, two expert sonologists performed, interpreted, and recorded videos of consecutive patients from September 2018 to April 2020. These videos captured the sliding sign test and provided the necessary data for training the DL model.

Training the DL Model

The DL model was trained using a dataset of TVS videos. A temporal residual network was used as the basis for the DL model. The videos were split into training, validation, and testing sets while maintaining a similar proportion of positive and negative sliding signs.

Evaluating the DL Model

The DL model was evaluated using an independent test set. The diagnostic accuracy of the model was compared to the reference standard sonologist classification. The following metrics were used to assess the model’s performance:
– Area under the receiver operating characteristic curve (AUC)
– Accuracy
– Sensitivity
– Specificity
– Positive and negative predictive value (PPV/NPV)

Results

The dataset consisted of 749 videos. Of these, 646 (86.2%) videos depicted a positive sliding sign, while 103 (13.8%) videos depicted a negative sliding sign. The DL model was applied to the test dataset using a threshold of 0.9.

The results were as follows:
– AUC: 96.5% (95% CI: 90.8-100.0%)
– Accuracy: 88.8% (95% CI: 83.5-92.8%)
– Sensitivity: 88.6% (95% CI: 83.0-92.9%)
– Specificity: 90.0% (95% CI: 68.3-98.8%)
– PPV: 98.7% (95% CI: 95.4-99.7%)
– NPV: 47.7% (95% CI: 36.8-58.2%)

Conclusion

In conclusion, we have successfully developed a DL model for the prediction of the TVS-based sliding sign classification. This model accurately determines the state of the POD, showing promising results in terms of accuracy, sensitivity, specificity, and predictive values. This AI-based solution has the potential to assist healthcare professionals in diagnosing and treating patients with endometriosis, specifically in identifying POD obliteration. By leveraging artificial intelligence, we can enhance the knowledge and understanding of this condition, ultimately improving patient outcomes. AI for Queensland businesses longtail.