Two prediction maps of slides that the CNN prognosticated correctly with high...
Two prediction maps of slides that the CNN prognosticated correctly with high probability are shown. Blue patches indicate a probability score < 0.5 and thus classified as 5y-OS(+) while red patches indicate a score > 0.5 and thus classified as 5y-OS(-). 5y-OS = 5-year overall survival. CNN = convolutional neural network. (A) The slide correctly classified as 5y-OS(-) shows a high-grade renal cell carcinoma with greatly enlarged, partly multinuclear nucleoli, heterogeneous nuclear atypia and accompanying inflammatory reaction. Scale bar = 100μm. (B) The slide correctly classified as 5y-OS(+) shows a low-grade renal cell carcinoma with still mostly uniform nuclei and no areas of a more aggressive type. Scale bar = 100μm.

Image source: Wessels et al., PLOS One 2022 (CC BY 4.0)

CNN applications in pathology

‘Deep learning can predict survival directly from histology in clear cell renal cell carcinoma’

Renal cell carcinoma is among the fifteen most common cancers worldwide. Surgical removal of the tumour is recommended for localized stages and oligometastatic cases, while non-resectable metastases require systemic therapy and optional cytoreductive nephrectomy.

portrait of Titus Brinker
Dr Titus Brinker

With clear cell renal cell carcinoma, risk-dependent diagnostic and therapeutic algorithms are routinely implemented in clinical practice. ‘Artificial intelligence-based image analysis has the potential to improve outcome prediction and thereby risk stratification. Thus, we looked at whether a convolutional neural network (CNN) can extract relevant image features from a typical H&E-stained slide to predict 5-year overall survival,’ said Dr Titus Brinker, head of Digital Biomarkers for Oncology group at the German Cancer Research Center (DKFZ) in Heidelberg. 

His CNN-based prognostication of overall survival demonstrated promising results in a study published in August, with a mean weighted accuracy of 72%, sensitivity of 72.4% and specificity of 71.7%. Its generalizability can be combined with existing clinical pathology parameters. ‘While this widely applicable technique shows the potential of AI in image-based outcome prediction, further research is needed to fine-tune this method and improve robustness,’ Brinker concludes.


Read all latest stories

Related articles


News • RNA sequencing

Using AI to establish molecular tumor classification

Treating physicians need information about the molecular subtype of the tumor if they are to provide targeted therapy for colorectal carcinoma. A research team from University Hospital Zurich and the…


Article • Renal carcinoma differentiation

SPECT/CT improves diagnosis of indeterminate kidney tumours

99mTc-Sestamibi SPECT/CT can aid in the diagnosis of solid renal lesions, especially in cases where contrast enhanced CT findings are inconclusive, according to researchers from the Mallinckrodt…


Article • AI provides prognostic information

Next-generation deep learning models predict cancer survival

Deaths from cancer are currently estimated at 10 million each year worldwide. Conventional cancer staging systems aim to categorize patients into different groups with distinct outcomes. ‘However,…

Related products

Arietta 650 DeepInsight


Fujifilm · Arietta 650 DeepInsight



Fujifilm · Dynamyx



Image Information Systems · iQ-VIEW

IMAGE Information Systems Europe GmbH
3DQuorum SmartSlices

Artificial Intelligence

Hologic · 3DQuorum SmartSlices

Hologic, Inc.
Advanced intelligent Clear-IQ Engine for MR

Artificial Intelligence

Canon · Advanced intelligent Clear-IQ Engine for MR

Canon Medical Systems Europe B.V.
Subscribe to Newsletter