Oral squamous cell carcinoma (OSCC) is a subset of head and neck squamous cell carcinoma (HNSCC), the 7th most common cancer worldwide, and accounts for more than 90% of oral malignancies. Early detection of OSCC is essential for effective treatment and reducing the mortality rate. However, the gold standard method of microscopy-based histopathological investigation is often challenging, time-consuming and relies on human expertise. Automated analysis of oral biopsy images can aid the histopathologists in performing a rapid and arguably more accurate diagnosis of OSCC. In this study, we present deep learning (DL) based auto- mated classification of 290 normal and 934 cancerous oral histopathological images published by Tabassum et al (Data in Brief, 2020). We utilized transfer learning approach by adapt- ing three pre-trained DL models to OSCC detection. VGG16, InceptionV3, and Resnet50 were fine-tuned individually and then used in concatenation as feature extractors. The con- catenated model outperformed the individual models and achieved 96.66% accuracy (95.16% precision, 98.33% recall, and 95.00% specificity) compared to 89.16% (VGG16), 94.16% (In- ceptionV3) and 90.83% (ResNet50). These results demonstrate that the concatenated model can effectively replace the use of a single DL architecture.

No abstract

Malaria is a lethal disease responsible for thousands of deaths worldwide every year. Manual methods of malaria diagnosis are timeconsuming that require a great deal of human expertise and efforts. Computerbased automated diagnosis of diseases is progressively becoming popular. Although deep learning models show high performance in the medical field, it demands a large volume of data for training which is hard to acquire for medical problems. Similarly, labeling of medical images can be done with the help of medical experts only. Several recent studies have utilized deep learning models to develop efficient malaria diagnostic system, which showed promising results. However, the most common problem with these models is that they need a large amount of data for training. This paper presents a computer-aided malaria diagnosis system that combines a semi-supervised generative adversarial network and transfer learning. The proposed model is trained in a semi-supervised manner and requires less training data than conventional deep learning models. Performance of the proposed model is evaluated on a publicly available dataset of blood smear images (with malariainfected and normal class) and achieved a classification accuracy of 96.6%.

Like many cancers, esophageal squamous cell carcinoma, has a substantially better survival rate if detected in early stages, making early detection a critical element in successful treatment. On the other hand, deep learning tools that employ high-resolution whole slide images and therefore substantially greater compute power and are not easily accessible in low-resource settings. In this study we train and test deep learning models using low-resolution and low-cost digital histopathology images acquired using a 5-megapixel Leica ICC50 camera mounted on a Leica DM500 binocular microscope, at two different magnification levels, Under IRB approvals, we acquired a total of 64 Hematoxylin and Eosin (H&E) stained tissue slides of biopsies taken from patient presenting at Rehman Medical Institute from 2016 to 2020 that were anonymized. A total of 2370 images were captured at two different magnification levels, i.e. 10x and 40x, and were labelled with four classes by at least one expert pathologist (677 well differentiated, 1066 moderately differentiated, 152 poor differentiated and 475 normal) to develop the ESCC image dataset. Preprocessing was performed to clean the dataset, generate patches and resize the images. Deep learning models were trained on the ESCC image dataset for feature extractions. This was followed by machine learning classifier for the diagnosis of ESCC. Deep learning models that were used in this study are Resnet50, VGG16, VGG19, and Inception_Resnet_V2 with Logistic Regression (LR) used as classifier. An ensemble model which stacks up the Inception and Resnet. This model preformed with the highest accuracy at 96% with Precision, Recall, and f1 score values at 96%, 96%, and 96%, respectively. This study reports 1) a local ESCC image repository for research purposes, 2) an ensemble ML/DL model that predicts ESCC with 96% accuracy and 3) also a computer-aided diagnosis (CAD) system for ESCC with a user-interface which will assist histopathologists in diagnosing ESCC with higher speed, efficiency and accuracy. Citation Format: Mansoor Alam, Ibrar Amin, Ayesha Sajjad, Hina Zamir, Saad Zaheer, Maria T. Khattak, Iqbal Muhammad, Faisal F. Khan. Using low-resolution, low-cost histopathology images to predict esophageal squamous cell carcinoma via deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 6329.