import gradio as gr import os from useful_functions import * import useful_functions from dotenv import load_dotenv load_dotenv() f_load_cancer_classifier() f_load_cnn_model() HF_TOKEN = os.getenv('HF_TOKEN') hf_writer = gr.HuggingFaceDatasetSaver(HF_TOKEN, "mole-dataset", private=True) def image_classifier(file_path, age, sex, localization): if age == 0: age = 40 if sex == "": sex = "unknown" if localization == "": localization = "unknown" # file_path = file_path if file_path is not None else file_path_webcam preds = f_predict_cnn_with_tta(file_path) label = f_predict_cancer(preds, age, sex, localization) return (dict(zip(useful_functions.lesion_model.dls.vocab, preds)), label) input_img = gr.Image(tool="editor", type="filepath", source="upload") # input_webcam = gr.Image(tool="editor", type="filepath", source="webcam") input_age = gr.Number(label="age (optionnel)") input_sex = gr.Dropdown(label="sex (optionnel)", choices=["male", "female"]) input_localization = gr.Dropdown(label="localization (optionnel)", choices=["abdomen", "back", "chest", "ear", "face", "foot", "genital", "hand", "lower extremity", "neck", "scalp", "trunk", "upper extremity"]) output_lesion = gr.Label(label="Lesion detected") output_malign = gr.Label(label="Classification") list_files_examples = os.listdir("examples") # examples = [[os.path.join("examples", file), 0, "", ""] for file in list_files_examples if file.endswith("jpg")] examples = [] examples.append([os.path.join("examples", "PXL_20221103_153018529.jpg"), 40, "female", "back"]) examples.append([os.path.join("examples", "PXL_20221103_153129579.jpg"), 40, "male", "neck"]) examples.append([os.path.join("examples", "PXL_20221103_153137616.jpg"), 40, "male", "neck"]) examples.append([os.path.join("examples", "PXL_20221103_153217034.jpg"), 40, "male", "back"]) examples.append([os.path.join("examples", "PXL_20221103_153256612.jpg"), 40, "male", "upper extremity"]) examples.append([os.path.join("examples", "ISIC_0025402.jpg"), 70, "male", "lower extremity"]) demo = gr.Interface(title="Skin mole analyzer", description=r"""This is a side project I have been working on to practice working with images. The purpose is to classify skin lesions (Based on kaggle dataset Skin Cancer MNIST: HAM10000). The framework used is FastAI/pytorch and the model used is a pre-trained cnn (resnet152). I added an extra layer to use age, sex, localization and output of resnet152 to classify the lesion as suspicious or not (randomForest model). The lesions detected are the following: This is in no case intended as a medical advice, just a pedagogical exercise.
*Pictures should be relatively well centered on the mole to obtain the best results (cf examples). You can use the tools available in the right corner to crop optimally. """, article="""[1] Noel Codella, Veronica Rotemberg, Philipp Tschandl, M. Emre Celebi, Stephen Dusza, David Gutman, Brian Helba, Aadi Kalloo, Konstantinos Liopyris, Michael Marchetti, Harald Kittler, Allan Halpern: "Skin Lesion Analysis Toward Melanoma Detection 2018: A Challenge Hosted by the International Skin Imaging Collaboration (ISIC)", 2018;""https://arxiv.org/abs/1902.03368"
[2] Tschandl, P., Rosendahl, C. & Kittler, H. The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Sci. Data 5, 180161 doi:10.1038/sdata.2018.161 (2018).""", fn=image_classifier, inputs= [input_img, input_age, input_sex, input_localization], outputs=[output_lesion, output_malign], examples=examples, allow_flagging="auto", flagging_options=list(useful_functions.lesion_model.dls.vocab) + ["other"], flagging_callback=hf_writer ) demo.launch()