Deep learning with transfer learning in pathology. Case study: classification of basal cell carcinoma

Abstract

Establishing basal cell carcinoma (BCC) subtype is sometimes challenging for pathologists. Deep-learning (DL) algorithms are an emerging approach in image classification due to their performance, accompanied by a new concept -transfer learning, which implies replacing the final layers of a trained network and retraining it for a new task, while keeping the weights from the imported layers. A DL convolution-based software, capable of classifying 10 subtypes of BCC, was designed. Transfer learning from three general-purpose image classification networks (AlexNet, GoogLeNet, and ResNet-18) was used. Three pathologists independently labeled 2249 patches. Ninety percent of data was used for training and 10% for testing on 100 independent training sequences. Each of the resulted networks independently labeled the whole dataset. Mean and standard deviation (SD) accuracy (ACC) [%]/sensitivity (SN) [%]/specificity (SP) [%]/area under the curve (AUC) for all the networks was 82.53±2.63/72.52±3.63/97.94±0.3/0.99. The software was validated on another 50-image dataset, and its results are comparable with the result of three pathologists in terms of agreement. All networks had similar classification accuracies, which demonstrated that they reached a maximum classification rate on the dataset. The software shows promising results, and with further development can be successfully used on histological images, assisting pathologists' diagnosis and teaching.

Improved artificial intelligence discrimination of minor histological populations by supplementing with color-adjusted images

“…Histopathological AI has surpassed first generation binary classification 21 , second generation cancer tissue classifications 22 , and is now in the third generation of tissue discrimination [23][24][25][26][27][28][29] . However, AI classification has not reached the capability of providing the detailed tissue classifications required in clinical practice.…”

“…Histopathological AI investigations targeting tissue subtypes have increased in recent years [23][24][25][26][27][28][29] . Concerning the composition of the training data in these papers, the mean percentages of training data with the most and least frequent tissue types were 30.3 ± 8.1% (22.6 -44%, 95% CI: 20.2 -40.4%) and 7.7 ± 6.2% (1.1 -16.2%, 95% CI: 0.0 -15.4%), respectively, with a disparity of 9.8 ± 10.1 fold (1.8 -21.2 fold, 95% CI: -2.6 -21.2 fold) [23][24][25][26][27][28][29] . Compared with those manuscripts, this study was capable of expanding this gap from a minimum of 1% to a maximum of 81.7%, ratios that are relatively close to the clinical epidemiology of thyroid cancer.…”

“…AI investigations were initiated as a binarization task for tumor classification between benign or malignant lesions 21 , followed by the rough discrimination of adenocarcinoma and squamous cell carcinoma 22 . At present, more detailed classification of tissue subtypes is being investigated [23][24][25][26][27][28][29] ; however, most efforts are limited to simply increasing the broad kinds of histologic subtypes. The development of immunohistochemical staining and genetic analyses have accelerated the tissue subdivision of malignant tumors.…”

Immunohistochemical expression of p53, Ki67, α-SMA, CD44 and CD31 in different histological subtypes of basal cell carcinoma

“…Perineural invasion may also be an indicator of aggressive disease [ 1 , 2 , 3 ]. BCC is a frequently diagnosed cancer with variable HP subtypes: nodular, the most common subtype; superficial, the most common subtype in younger age groups; morpheaform, the biological behavior is more aggressive; basosquamous, histological features of both BCC and SCC; micronodular, destructive behavior and high rates of recurrence and adenoid [ 4 , 5 ]. Exposure to ultraviolet radiation (UVR) can cause mutations in the p53 gene, which is the most frequent genetic abnormality in skin cancers.…”

Nodular and Micronodular Basal Cell Carcinoma Subtypes Are Different Tumors Based on Their Morphological Architecture and Their Interaction with the Surrounding Stroma

“…In a previous study, the present group of authors designed a deep learning convolution-based software using transfer learning from three general-purpose image classification networks: AlexNet, GoogLeNet and ResNet-18. This software was able to classify subtypes of BCC, such as superficial, nodular (with adenoid, nodulo-cystic and keratotic variants), pigmented, with adnexal differentiation, micronodular and infiltrating [ 24 ].…”