Skip to main content

Table 1 Glance of deep learning-based lung cancer histological classification algorithms and general slide image analysing tools

From: Deep learning-based six-type classifier for lung cancer and mimics from histopathological whole slide images: a retrospective study

Research

Year

Objective

Cohort

AUC

Architecture

Framework

Language

Coudray et al. [16]

2018

Classification between LUAD, LUSC, and NL; mutation prediction (STK11, EGFR, FAT1, SETBP1, KRAS, and TP53)

TCGA (1634 slides); NYU (340 slides)

0.970 (classification) 0.733–0.856 (mutation)

Inception-V3

TensorFlow

Python

Yu et al. [17]

2020

Identification of histological types and gene expression subtypes of NSCLC

ICGC (87 LUAD patients, 38 LUSC patients); TCGA (427 LUAD patients, 457 LUSC patients)

0.726–0.864

AlexNet; GoogLeNet; VGGNet-16; ResNet-50

Caffe

Python

Gertych et al. [18]

2019

Histologic subclassification of LUAD (5 types)

CSMC (50 cases); MIMW (38 cases); TCGA (27 cases)

Accuracy, 0.892 (patch-level)

GoogLeNet; ResNet-50; AlexNet

Caffe

MATLAB

Wei et al. [19]

2019

Histologic subclassification of LUAD (6 types)

DHMC (422 LUAD slides)

0.986 (patch-level)

ResNet-18

PyTorch

Python

Kriegsmann et al. [20]

2020

Classification between LUAD, LUSC, SCLC and NL

80 LUAD, 80 LUSC, 80 SCLC and 30 controls from NCT

1.000 (after strict QC)

Inception-V3

Keras (TensorFlow)

R

Wang et al. [21]

2020

Classification between LUAD, LUSC, SCLC, and NL

SUCC (390 LUAD; 361 LUSC; 120 SCLC; and 68 NL slides); TCGA (250 LUAD and 250 LUSC slides in good quality)

0.856 (for TCGA cohort)

Modified VGG-16

TensorFlow

Python

QuPath [22]

2017

Tumour identification, biomarker evaluation, batch-processing, and scripting

Specimens of 660 stage II/III colon adenocarcinoma patients from NIB

/

/

/

JAVA

DeepFocus [23]

2018

Detection of out-of-focus regions in WSIs

24 slides from OSU

/

CNN

TensorFlow

Python

ConvPath [24]

2019

Cell type classification and TME analysis

TCGA (LUAD); NLST; SPORE; CHCAMS

/

CNN

/

MATLAB; R

HistoQC [25]

2019

Digitization of tissue slides

TCGA (450 slides)

/

/

/

HTML5

ACD model [26]

2015

Colour normalization for H&E-stained WSIs

Camelyon-16 (400 slides); Camelyon-17 (1000 slides); Motic-cervix (47 slides); and Motic-lung (39 slides)

0.914 (for classification)

ACD

TensorFlow

Python

  1. Abbreviations: LUAD, lung adenocarcinoma; LUSC, lung squamous cell cancer; NL, normal lung; TCGA, the Cancer Genome Atlas; NYU, New York University; ICGC, International Cancer Genome Consortium; CSMC, Cedars-Sinai Medical Center; MIMW, Military Institute of Medicine in Warsaw; DHMC, Dartmouth-Hitchcock Medical Center; NCT, National Center for Tumor Diseases; QC, quality control; SUCC, Sun Yat-sen University Cancer Center; NIB, Northern Ireland Biobank; OSU, Ohio State University; NLST, National Lung Screening Trial; SPORE, Special Program of Research Excellence; CHCAMS, Cancer Hospital of Chinese Academy of Medical Sciences; H&E, haematoxylin and eosin; WSIs, whole slide images; ACD, adaptive colour deconvolution