MOC31 Immunostaining in the Diagnosis of Metastatic Adenocarcinoma in Serous Fluid: Special Emphasis on Atypical Cytological Cases

Using a deep learning neural network for the identification of malignant cells in effusion cytology material

AIM

To evaluate the application of an artificial neural network in the detection of malignant cells in effusion samples.

MATERIALS AND METHODS

In this retrospective study, we selected 90 cases of effusion cytology samples over 2 years. There were 52 cases of metastatic adenocarcinoma and 38 benign effusion samples. In each case, an average of five microphotographs from the representative areas were taken at 40× magnification from Papanicolaou-stained samples. A total of 492 images were obtained from these 90 cases. We applied a deep convolutional neural network (DCNN) model to identify malignant cells in the cytology images of effusion cytology smears. The training was performed for 15 epochs. The model consisted of 783 layers with 188 convolution-max pool layers in between.

RESULTS

In the test set, the DCNN model correctly identified 54 of 56 images of benign samples and 49 out of 56 images of malignant samples. It showed 88% sensitivity, 96% specificity and 96% positive predictive value in the screening of malignant cases in effusion. The area under the receiver operating curve was 0.92.

CONCLUSION

DCNN is a unique technology that can detect malignant cells from cytological images. The model works rapidly and there is no bias in cell selection or feature extraction. The present DCNN model is promising and can have a significant impact on the diagnosis of malignancy in cytology.

Serous cavity metastasis: Evaluation of unknown primary

Malignant effusions can occur in patients with neoplasia. Once a metastatic diagnosis is confirmed, the primary site of origin of malignancy needs to be ascertained. This task can be challenging without a prior history of malignancy. In some patients their effusion may be the initial presentation of an underlying malignancy. Metastases usually present with a dual population of mesothelial and malignant cells. Combining cytomorphologic examination with ancillary testing such as immunocytochemistry can help identify the origin of the foreign malignant cell population. Helpful architectural clues include a single cell pattern, solid cell ball pattern, single file arrangement, papillary formation, psammoma bodies and background mucin. Useful cellular features include the presence of signet ring cells, small cells, pleomorphic and multinucleated giant cells, squamous cells, spindle cells and pigmentation. Rarely, despite an extensive work-up the primary site of origin for a malignant effusion may remain unresolved. This review article will be incorporated finally as one of the chapters in CMAS (CytoJournal Monograph/Atlas Series) #2. It is modified slightly from the chapter by the initial authors in the first edition of Cytopathologic Diagnosis of Serous Fluids.