A New Kind of Breast Imaging

Radiomics uses imaging data as information carriers. In addition to the conventional morphological pictures calculated from data, it extracts data patterns.

„There is a lot of mystification about radiomics – many people write about it, but not everybody knows what it is,“ said Lale Umutlu, Essen University, Germany. She gave the definition right away: Radiomics is the conversion of images to higher dimensional data.
“We are trying to gain added value to imaging,” explained Umutlu, using lung cancer as an example. While conventional radiology may describe a lesion in the lower lobe of the right lung, radiomics will add parameters like the lesion’s geometry, density, contrast pattern and many more features.

How Does Radiomics Work?

The first part is about finding the right data. Radiomics starts with imaging data, which is segmented based on pathology. From the segments, “hundreds and thousands of features are extracted”, said Umutlu. Using expert opinion to narrow down the most likely features is a common approach. However, no standards for finding the right parameters have been established yet.

The data is then analyzed with various mathematical models. This part of the process is “a very large black box,” said Umutlu later in the discussion.
The second part is about validation. Internal validation and cross validation are followed by external validation. “External validation is quite complicated,” noted Umutlu. It checks whether a model works in different centers and regardless of particular vendors or machines.

Umutlu showed a successful radiomics study from her institution about KRAS (Kirsten rat sarcoma viral oncogene homolog), which is one of the most frequently mutated genes in non–small cell lung cancer (NSCLC).
Baseline CTs of 53 patients were analyzed using 1829 features. The analysis showed that patients could be subdivided into two cohorts: A high-risk cohort with an overall survival of ten months, and a low-risk cohort surviving 21.3 months. The stratification was solely based on imaging data extracted from the baseline CTs.

Umutlu presented another study from Essen University, including 50 patients with NSCLC stage IV: This time, clinical features like tumor burden, hemoglobin and C-reactive protein (CRP) worked together with baseline CT data (1773 features were used) to better stratify patients who would benefit form immunotherapy. “In this case, radiomics clearly uplifted our diagnostic power for this group of patients,” said Umutlu.

Radiomics in Breast Imaging

Radiomics in breast imaging is at the very start. “We currently have no more than forty to forty-five quality publications,” said Umutlu. She introduced a few:

  • Parekh (2017) found that radiomics based on T1w, T2w, dynamic contrast-enhanced MRI and DWI better differentiated benign from malignant tumors.
  • Bickelhaupt (2017) also based the differentiation between benign and malignant on DWI and 1.5T MRI. False positive findings significantly decreased. Especially BIRADS 4a and 4b lesions benefited from the method.
  • Dong (2018) also found a marked improvement for differentiating malignant form benign breast tissue.
  • Li (2016) found that radiomics features can be associated with recurrence.
  • Guo (2017) used radiomics on ultrasound features. Various parameters, especially the echo pattern of invasive ductal carcinoma. Hormone receptor positive and EGFR2-negative cancers differ from triple negative cancers.


Radiomics relies on a completely new approach to imaging data. “We will build something new, but it will take another five to ten years to find out which part of this hype is worth it”, said Umutlu. External validation in other institutions and on different scanners is the key to further rolling out radiomics, said Umutlu: “It is not easy and everybody is staring with internal validation, but external validation is what we really need.”


Bickelhaupt S et al. Prediction of malignancy by a radiomic signature from contrast agent-free diffusion MRI in suspicious breast lesions found on screening mammography. J Magn Reson Imaging. 2017;46:604-16.

Dong Y et al. Preoperative prediction of sentinel lymph node metastasis in breast cancer based on radiomics of T2-weighted fat-suppression and diffusion-weighted MRI. Eur Radiol. 2018;28:582-91.

Guo Y et al. Radiomics Analysis on Ultrasound for Prediction of Biologic Behavior in Breast Invasive Ductal Carcinoma. Clin Breast Cancer. 2017 Aug 18. pii: S1526-8209(17)30146-5.

Li H. MR Imaging Radiomics Signatures for Predicting the Risk of Breast Cancer Recurrence as Given by Research Versions of MammaPrint, Oncotype DX, and PAM50 Gene Assays. Radiology. 2016 Nov;281(2):382-91.

Parekh VS, Jacobs MA. Integrated radiomic framework for breast cancer and tumor biology using advanced machine learning and multiparametric MRI. NPJ Breast Cancer. 2017 Nov 14;3:43.