Classification of subtypes including LCNEC in lung cancer biopsy slides using convolutional neural network from scratch

Abstract

Identifying the lung carcinoma subtype in small biopsy specimens is an important part of determining a suitable treatment plan but is often challenging without the help of special and/or immunohistochemical stains. Pathology image analysis that tackles this issue would be helpful for diagnoses and subtyping of lung carcinoma. In this study, we developed AI models to classify multinomial patterns of lung carcinoma; ADC, LCNEC, SCC, SCLC, and non-neoplastic lung tissue based on convolutional neural networks (CNN or ConvNet). Four CNNs that were pre-trained using transfer learning and one CNN built from scratch were used to classify patch images from pathology whole-slide images (WSIs). We first evaluated the diagnostic performance of each model in the test sets. The Xception model and the CNN built from scratch both achieved the highest performance with a macro average AUC of 0.90. The CNN built from scratch model obtained a macro average AUC of 0.97 on the dataset of four classes excluding LCNEC, and 0.95 on the dataset of three subtypes of lung carcinomas; NSCLC, SCLC, and non-tumor, respectively. Of particular note is that the relatively simple CNN built from scratch may be an approach for pathological image analysis.

Deep Learning for Lung Cancer Diagnosis, Prognosis and Prediction Using Histological and Cytological Images: A Systematic Review

“…The higher performance was in the study of Kanavati et al [24] (AUC of 0.94-0.99), which included a large number of images. Only two studies designed a classification task for identifying ADC, SCC, SCLC, and LCNEC on WSIs [27,28]. This 4-class task represents the realistic daily practice of a pathologist.…”

“…The InceptionV3 model achieved the highest performance; however, many cases of ADC and SCC were misclassified [26]. In a retrospective study by Yang et al, a six-type classifier model was designed for lung cancer (ADC, SCC, SCLC) as well as other lung diseases (pulmonary tuberculosis, organizing pneumonia) subtyping on H&E-stained slides [27]. The proposed classification task achieved great performance and consistency with experienced pathologists.…”

“…Apart from these, U-Net [116,119], Xception [58,94,96], Hover-Net [76,83] and InceptionV3 [41] have also been used in several studies. In comparative studies or studies which use multiple architectures of CNNs, several other models have been presented, such as NASNetLarge [27], EfficientNet [26], SqueezeNet [39], etc. On the other hand, few approaches utilized known DL detectors for segmentation or quantitation purposes.…”

Current status and prospect of PET-related imaging radiomics in lung cancer

“…AI-based radiomics involves extracting quantitative features from extensive medical images and constructing predictive models that correlate image features with clinical endpoints, all of which fall under supervised learning. The radiomics workflow can be divided into five parts (as depicted in Figure 1): [1] image labeling, [2] feature extraction, [3] feature selection, [4] model construction, and [5] performance evaluation. Traditional ML algorithms are involved in [2] through [5], while the DL strategy primarily plays a role in [1] and [2], offering potential to enhance the automation and efficiency of radiomics analysis.…”

“…Stage IIIA~IVA patients have survival rates as low as 10%~36%, whereas stage I patients may achieve survival rates ranging from 77% to 92% ( 3 ). Consequently, early detection emerges as the crucial factor in reducing patient mortality ( 1 – 4 ). Although there has been some progress in lung cancer survival rates in recent decades, the overall 5-year survival rate remains low, typically ranging between 10% and 20% ( 2 ).…”