Machine learning fusion prototype could potentially help detect ovarian cancer
Ovarian cancer, or cancer in the ovaries, is one of the deadliest form of cancer for women. However, it is often detected at later stages. Only approximately 20% of ovarian cancer cases are found at an early stage. Unfortunately, there are no actual screening tests for ovarian lesions. Also, ovarian lesions are difficult to diagnose accurately. This has led to 80% of patients having no sign of cancer in their ovaries, only to undergo surgery and have lesions removed and tested.
Researchers at Washington University in St. Louis’ McKelvey School of Engineering in the U.S. have applied various imaging methods to diagnose ovarian cancer more precisely and accurately.
The study was published in the journal Photoacoustics.
A new machine learning prototype
Quing Zhu, the Edwin H. Murty Professor of Biomedical Engineering at Washington University in St. Louis’ McKelvey School of Engineering, and her research team have developed a new machine learning (ML) fusion model that uses existing ultrasound features of ovarian lesions to train and ‘teach’ the model to recognize whether a lesion is benign or cancerous. A machine learning fusion model uses numerous machine learning algorithms. It then combines them all to complete a task, solve a problem or make a prediction.

The method utilizes reconstructed images taken with photoacoustic tomography. This type of technique, also called optoacoustic tomography, allows researchers to take 3D images of a specific area, providing spatial resolution and depth of the dimension.
Multi-modality machine learning
Machine learning has typically been focused on single modality data, or a classification of input and output between a computer and a person, otherwise known as a dataset. Recent findings suggest that multi-modality machine learning is more detailed and accurate in its performance over single modality machine learning methods. In a preliminary study of 35 patients with over 600 regions of interest, the model’s accuracy was 90%.
More information can be identified using photoacoustic imaging
The study is the first to use ultrasound to improve the machine learning performance of photoacoustic tomography reconstruction for cancer diagnosis. The photoacoustic imaging uses light and sound to detect cancer in the ovaries. “Existing modalities are mainly based on the size and shape of the ovarian lesions, which do not provide an accurate diagnosis for earlier ovarian cancer and for risk assessment of large adnexal/ovarian lesions,” said Quing Zhu. “Photoacoustic imaging adds more functional information about vascular contrast from hemoglobin concentration and blood oxygen saturation.”
The creation of the novel method for diagnosing ovarian cancer
Yun Zou, a researcher on the team and a student in Zhu’s lab, developed the new machine learning fusion model by combining an ultrasound neural network with a photoacoustic tomography neural network to perform the ovarian lesion diagnosis. The cancerous lesions in the ovaries can be seen in the imaging from the ultrasound, either as a solid lesion or with papillary projections, or tumorous growths from tissue that line the organ, inside the cystic lesions. The team also included the total hemoglobin concentration and blood oxygenation saturation from the imaging, which are two biomarkers of ovarian cancer. This would help improve the accuracy of diagnosis using ultrasound.
“Our results showed that the ultrasound-enhanced photoacoustic imaging fusion model reconstructed the target’s total hemoglobin and blood oxygen saturation maps more accurately than other methods and provided an improved diagnosis of ovarian cancers from benign lesions,” Zou stated.
The researchers want to use this study to create a screening tool that would allow for earlier diagnosis of ovarian cancer, detecting the disease before it reaches its later stages.